Canon Unveils 120-Megapixel Camera Sensor

Barence writes "Canon claims to have developed a digital camera sensor with a staggering 120-megapixel resolution. The APS-H sensor — which is the same type that is used in Canon's professional EOS-1D cameras — boasts a ridiculous resolution of 13,280 x 9,184 pixels. The CMOS sensor is so densely packed with pixels that it can capture full HD video on just one-sixtieth of the total surface area. However, don't hold your breath waiting for this baby to arrive in a camera. Canon unveiled a 50-megapixel sensor in 2007, but that's not made it any further than the labs to date." It's probably not going too far out on a limb to say that the any-day-now rumored announcement of an update to the 1D won't include this chip, but such insane resolution opens up a lot of amazing possibilities, from cropping to cheap telephoto to medium and large format substitution. Maybe I should stop fantasizing about owning a full-frame 1D or 5D and redirect my lust towards 120 megapixels.

Comments

Noise/Light Sensitivity/Optics

[Greymist]

I'm just curious what this would be like in low light settings, cramming that many pixels into such a small space has got to have some effect on sensitivity.

Re:Noise/Light Sensitivity/Optics

[spun]

I'd bet that you could use that many megapixels to seriously boost dynamic range by averaging several adjacent pixels into one.

Re:Noise/Light Sensitivity/Optics

[wjh31]

the pixel density is about that of a modern compact, back of the envelope calulation suggests about the same as 10Mpixels on a 1/2.5" sensor. So i guess low light performance would be comparable to modern compacts.

Re:Noise/Light Sensitivity/Optics

[bieber]

Well, its likely application is in controlled commercial use, medical imaging and the like I would imagine, so any near-term use will almost certainly be under controlled lighting conditions.

That being said, give 'em five years and they'll probably have it in a 1D churning out noiseless photos at ISO 51200 or some such nonsense. I used to think my 20D's marginally-usable ISO 3200 was pretty darn impressive, and now we've got insanely high res cameras doing 12800 and still looking decent. The way sensor technology keeps improving, I'm not even going to try predicting how far they may get even in the next couple of years, let alone decades down the road...

Re:Noise/Light Sensitivity/Optics

[john83]

It's an interesting question. Pixels return an intensity value proportional to the mean intensity over their surface, so I'd imagine you could average groups of 2x2 or 3x3 (etc.) pixels to trade resolution for sensitivity. Alternatively, you could up the gain on each pixel, which as Greymist points out would reduce your signal to noise ratio.

Re:Noise/Light Sensitivity/Optics

[crgrace]

I'm just curious what this would be like in low light settings, cramming that many pixels into such a small space has got to have some effect on sensitivity.

Pixel size, per se, has no impact on the light sensitivity of the pixel. That depends only on the read noise of the sensor and its associated electronics. A small pixel, however, does limit the depth of the potential well, so it would have more of an impact on in bright settings. What I'm saying is it would reduce the dynamic range of the sensor, but not have any direct effect on its performance in low light.

To get back the bright performance, pixels can be ganged together to make superpixels, but, of course, this trades a bit of resolution.

Re:Noise/Light Sensitivity/Optics

[Entropius]

Not quite. It's the same PER PIXEL: if you crop a 10MP area out of this thing then it'll be roughly as noisy as a 10MP 1/1.8" sensor. (Did you use crop factor 1.6 or 1.3 in your math? This sensor is APS-H, 1.3x, not the commonly-used 1.6x APS-C.) But if you print at any given output size, the pixels from the higher-resolution sensor will be smaller and thus whatever noise is present at a pixel level will be less intrusive.

For the math geeks: the real thing you should look at is the signal-to-noise ratio at any given spatial frequency (in cycles per image height).

Re:Noise/Light Sensitivity/Optics

[Entropius]

What's improving is in part processing technology. At some point you run into the brick wall of Planck's constant: the fluctuations in a counting experiment counting N photons can be no less than sqrt(N) due to shot noise.

Re:Noise/Light Sensitivity/Optics

[ceoyoyo]

How would that help dynamic range?

Re:Noise/Light Sensitivity/Optics

[crgrace]

Alternatively, you could up the gain on each pixel, which as Greymist points out would reduce your signal to noise ratio.

Actually, increasing the pixel gain *improves* the SNR. This is because the noise limitation of these sensors is virtually always the readout electronics. Therefore, adding as much gain as possible before the signal hits the readout chain will lower the overall noise of the system. This is analogous to using a low noise amplifier (LNA) in front of an RF receiver.

There are, of course, limitations. Pixels generally have a voltage gain of less than one (that is, the gain from the photodiode to the pixel output) so there isn't much you can do there.

Re:Noise/Light Sensitivity/Optics

[spun]

You may find this article helpful: http://en.wikipedia.org/wiki/High_dynamic_range_imaging

Re:Noise/Light Sensitivity/Optics

[idontgno]

So. Um.

You'd restore the dynamic range capability of this sensor to the level of lower-resolution (larger pixel) CCDs by... combining pixels? So you're back to lower-resolution imaging.

Are we being "Whooshed!" here? Or are you sincerely saying "Well, we have a 120 megapixel imager, but in order to get good dynamic range we have to process it back to 10 megapixels, just like your crappy cell phone camera."

Re:Noise/Light Sensitivity/Optics

[mangu]

Or are you sincerely saying "Well, we have a 120 megapixel imager, but in order to get good dynamic range we have to process it back to 10 megapixels, just like your crappy cell phone camera."

He's saying we have a 120 megapixel imager with great dynamic range (much, much better than any cell phone camera) and we can process it back to 10 megapixels to get awesome dynamic range.

Considering how big this chip is, even at 120 megapixels there's much more light gathering surface per pixel than in a shitty phone camera.

Re:Noise/Light Sensitivity/Optics

[ceoyoyo]

I'm familiar with HDR, thanks. You'll note that the article you linked to doesn't contain the words "average" or "averaging."

HDR requires that you have the same picture but with multiple, different exposures. You could potentially acquired this in one shot by making adjacent pixels more or less light sensitive (which has to be done in hardware), but averaging identical pixels isn't going to help. Nor does the HDR process involve averaging, even with multiple exposures.

Re:Noise/Light Sensitivity/Optics

[Jarik C-Bol]

not *quite* you could still get say, 40 megapixels. A very basic HDR picture is the combination of 3 ranges, so if you took your base picture on one pixel, and the bracket range on the pixels to its left and right, (i'm generalizing here of course, the tech would not be THAT simple) the output is a 40MP picture with a dynamic range 3x what you would get with a standard 40MP camera. the fact that your saying "get good dynamic range" shows that you don't know much about the subject. Normal cameras simply *don't* have dynamic range, they take photos of a very narrow range of light, based on your film shutter speed, etc. basically, a camera that automatically got ANY dynamic range, is an improvement towards being able to capture true to eye/life images.

Re:Noise/Light Sensitivity/Optics

[DrgnDancer]

I think what he's saying is, "We have this incredibly sensitive sensor array, in good light it can do 13,280 x 9,184 pixels; or, if the light isn't that good we can cut it to 6640 x 4592 and combine pixels to get a more accurate image despite the lack of light." It's like anything else, when conditions aren't ideal you lose stuff. In this case if your light isn't good you can (in theory) go from phenomenally high resolution to merely really high resolution by combining pixels. You're only getting 2.5 times as high a resolution as the best current cameras on the market instead of 5 times, but hopefully the image will be better.

Even operating at a quarter of its normal capability this sensor is a little better than the best currently on the market, so if it could (again theoretically) produce a more accurate (if lower resolution) image in low light situation, while producing a higher resolution image in normal light situations, that's a pretty good compromise.

Re:Noise/Light Sensitivity/Optics

[spun]

I'm saying, you could have your choice, more megapixels or more dynamic range, with the flip of a switch.

Re:Noise/Light Sensitivity/Optics

[Anonymous Coward]

I would rather have a lower megapixel sensor that works awesome in low light and high speed situations than a high mp sensor that sucks. Particularly for the types of situations 95% of the picture taking world would encounter.

A fine example would be comparing my point and shoot cameras:

One, a 3.2mp Canon A400 Circa 2004 built before the pixle wars. You have to look closely on the box to even figure out what mp it boasts and I'm not even sure it is posted on the camera at all. This camera takes OUTSTANDING pictures without fiddling with settings for a low-end point and shoot camera. The large (for a PnS) lense certainly helps.

Next, a Waterproof Pentax Optio WP30 at 7.1mp. Takes decent pictures. The small lense hurts a lot, and makes sensor sensitivity that much more important.

Finally, a Fuji Z33 WP at 10mp. Also a small lense. The pictures are so-so. Definitely not as good as the Pentax and no where near the quality of the Canon.

Granted, this is fairly anecdotal. Comparing one of the top camera brands with so-so brands that also have tiny lenses. but none the less. Why are my Megapixes going up and my quality going down? I'm convinced if my Pentax or Fuji cameras had the same image sensor as the Canon, their low light performance would improve greatly.

Re:Noise/Light Sensitivity/Optics

[spitzak]

Lots of people trying to correct you but I will add my own:

Averaging several adjacent pixels will increase the dynamic *resolution*. It will not increase the dynamic *range*.

If you have 100 samples between the range 0 and 1 and you average them together, the result is still in the 0 to 1 range.

Re:Noise/Light Sensitivity/Optics

[Beardydog]

This is an actual question directed at you, not an argument, so bear with me... In a frame that captures the full available range in a scene (where a bright sky, for example will have detail), the dark areas will be underexposed and noisy, but not completely black (the way an overexposed sky will appear completely white). Couldn't four adjacent pixels simply be added together to produce an image with four times the range, and one quarter the resolution? So if, for example, three of the underexposed pixels aren't lucky enough to get a single bit of light, and one pixel gets lucky and grabs some, you can treat it like a single pixel that's sensitive to units of light 1/4 as strong (abundant? I don't know how it works) as the actual pixels. You'd probably get extra noise and blurring added, (for pixels that sit on actual boundaries in the captured scene, but are merged together), but in principle, wouldn't you get more range information in such an image?

Re:Noise/Light Sensitivity/Optics

[vadim_t]

I don't think that would really work.

Think of a sensor like a bucket. Let's say it has a capacity for 256 drops. If your scene is lit in such a way that some buckets would overflow while others had just a few drops in them, that's when you have a dynamic range problem. You solve that by either having buckets that can hold more drops (which is why a DSLR has much more range than a cell phone), or by taking multiple exposures (HDR).

But I don't think you can post-process a high resolution image into a lower one with more dynamic range. Think of actual buckets. If you take a very shallow bucket and leave it in a storm for a minute to try to measure the amount of rain, and it overflows, then getting 3 buckets more of the same size isn't going to help you any. You'll just have 4 buckets that will overflow in the same amount of time. Connecting the buckets together won't help you either, as you still can store the same amount of water per square meter as before.

Re:Noise/Light Sensitivity/Optics

[spun]

Lots of people saying I'm right, too.

Dynamic resolution and dynamic range are the same thing. If you take the value of one pixel, it will be three integers. If you average the value of several adjacent pixels, you will have three reals. There are more real numbers between 0 and 255 than there are integers between 0 and 255, therefore, the range of values has increased. (0,0,0) is still pure black, and (255,255,255) is still white, you can't get any blacker than black or whiter than white, you know. But using reals, you have more values between black and white than you did, and therefore, more dynamic range.

Looked at another way, lets say a pixel is almost zero, or black. Using one pixel integers, it would round down to black, but averaging more than one pixel, one might find it wasn't quite black anymore. We have something between zero and one, i.e. greater dynamic range.

Re:Noise/Light Sensitivity/Optics

[rrohbeck]

You can crank up the sensitivity all the way and then run a low pass filter to get rid of all the noise ;)

Re:Noise/Light Sensitivity/Optics

[Zeety]

I wonder if we could have a have a LCD shutter on each pixel, exposing the element for a % of the total exposure time. That way we don't need to bake it in.

Re:Noise/Light Sensitivity/Optics

[Beardydog]

I've started taking multiple photos of everything I care about with my 5.0 Mp A610 and stitching them together. Even crushing the stitched image back down to the camera's resolution leaves you with a much better image than than straight shot. It's depressing.

As a side note, for anyone who hasn't stumbled across it: http://chdk.wikia.com/wiki/CHDK
I'm not sure if this is the original, or one-and-only, but it's possible to add RAW and other features (brickout : ) to cameras that don't normally support them with a little futzing around.

Re:Noise/Light Sensitivity/Optics

[mangu]

In strong light situations you either close your diaphragm letting less light in or do a shorter exposure. The problem is with low light situations.

In your bucket analogy, there will be a situation when only one drop will fall, it will hit one of the four buckets. Averaging them you have 0.25 drops in each of the buckets, but looking at the four buckets separately you will have one bucket with a pixel and three with none, i.e. a noisy picture.

Re:Noise/Light Sensitivity/Optics

[Threni]

They don't even have a colour for each pixel yet - just a monochrome sensor and a plastic coating to prevent/allow light of given wavelengths to hit each pixel, then this mess is converted into rgb pixels using some algorithm or other. This is why the number of megapixels for a given sensor is a bit of a joke - there are far fewer actual rgb pixels than there are monochrome pixels.

Re:Noise/Light Sensitivity/Optics

[danpbrowning]

Actually, the "sensitivity" (more specifically, e-/m^2) is generally the same across a huge variety of pixel sizes, thanks to microlenses. What is not usually constant is read noise (AKA "high ISO noise", sometimes also referred to as "sensitivity"), because although it does naturally shrink a little bit as the pixel size is reduced, it's not always in exact linear proportion with pixel diameter, and therefore the generalization that smaller pixels tend to have slightly more noise in low light.

Re:Noise/Light Sensitivity/Optics

[bws111]

Resolution and range are not the same thing. Resolution is the number of increments within the range. Range defines how dark your darkest area can be compared to how bright the brightest area can be. Resolution is the number of shades of grey between black and white. If you have some areas of the picture that exceed the blackest black and others that exceed the whitest white you don't have enough range, and averaging pixels can not correct that.

Re:Noise/Light Sensitivity/Optics

[danpbrowning]

Dynamic range is the distance between clipping and noise. The standard engineering definition assumes a SNR of 1:1 as the lower bound, but few photographers can tolerate that much noise and usually prefer 4:1 or 8:1. Random noise sources add in quadrature so that downsampling the pixel count by a factor of four increases the SNR by a factor of 2. A better way of thinking about it is this: the raw data from an image sensor has a noise power that increases linearly with spatial frequency (level of detail). Higher frequency (smaller details) have higher noise powers. If you throw away the high frequency detail, the noise goes with it. In actual practice, there are many better ways to reduce noise than by throwing away detail, and in any case, many viewers will prefer a detailed but noisey image over a blurry but less noisy one.

Re:Noise/Light Sensitivity/Optics

[Zeety]

Well the monochrome isn't really an issue here. I think we could get by with a monochrome LCD matrix sitting on top. I just dunno if they make them with that small of elements to do a one 2 one match with the CMOS.

Re:Noise/Light Sensitivity/Optics

[Threni]

You're forgetting that pixels on most digital cameras are monochrome, and there's a system of rgb filters over those sensors to get the colour, and an algorithm to turn a bunch of those rgb values into a single value for each pixel. Ie an 8 by 8 area of sensor isn't really 64 rgb pixels; it's 64/3 (roughly) of each of r,g and b components and this is turned into 64 pixels using interpolation.

You're making a mess of dynamic range though, too. For one thing, 255 doesn't come into it - you tend to get more than an 8 bit range. It's crummy jpegs which are stuck in the dark ages. Most reasonable cameras let you take pictures in RAW format, meaning you get closer to what the sensor recorded, without having lossless compress, 8 bit quantization or white balance etc burned into the image.

And even if your 255 is just an example, you don't make sense. Once you start averaging pixels (beyond what I referred to in my first paragraph) you're losing sharpness. And what are you going to do with your numbers, anyway? You have to convert them back to 0 -> 255 to store them in a jpeg.

Dynamic range is simply the fact in a room with dark dark bits,and bright bright bits, you have to choose what to expose. For example, in a dark church with no lighting, except for small stained glass windows with light streaming through them - if you adjust the settings to take a picture of the windows, you'll probably have quite a short shutter speed, or a small aperture, or a low ISO. Do this, and the dark parts of the photo will be very dark. You can try and make the dark bits bright in software, but you'll not have much data to go on, so you'll be adding a lot of noise.

If you try and expose for the dark parts by using a slow shutter, wide aperture of high iso, you're going to get a lot of dark detail, but you'll probably blow the highlights on the stained glass window. IE you'll exceed the maximum amount of light you can capture and end up with pure white. That's why professional photographers use filters when they do landscape shots - the top half will have a physical filter preventing as much light reaching the sensor as the bottom half. Some people use HDR to try and fix this - they take several identical shots on a tripod which differ only by shutter speed, and then try and combine the detail from the shots - take the bright bits from the fast shot and the dark detail from the lower shots. Personally, I think 99% of HDR shots look dreadful, because they are so utterly unrealistic, but they can be striking and it is possible to use it subtly, and I'm sure one day the software will be easier to use (that is, it'll be easier to make it look accurate - it's pretty easy now if you don't mind the results looking awful).

Re:Noise/Light Sensitivity/Optics

[vadim_t]

No, it's just as noisy.

You have 0.25 drops per 4 bucket group average, or 1 bucket with 1 drop and 3 with none, which averages to 0.25. Both images are really equivalent. You could scale down the second one, or scale up the first one and get the same result.

And none of those options will result in a good picture.

First reason is that if you're shooting in the darkness, there's got to be a source of light somewhere. If you shoot with a slow enough exposure to get some details in the shadows, it's nearly certain that something will get blown out. For instance, any light sources, and those are usually present on night pictures. You'll have street lights, car lights, establishment lighting, or something of the sort in the picture. Unless you shoot a landscape in the moonlight the light levels are pretty much guaranteed to be very uneven. And to deal with that you need good dynamic range.

The second reason is that the sensor is noisy by itself, and that makes the bottom of the bucket a somewhat fuzzy thing. You'll also have to amplify that to see it at all, and with a low dynamic range every light will turn into an ugly sphere of pure white, while dark details will have a lot of noise.

Re:Noise/Light Sensitivity/Optics

[Anonymous Coward]

Looked at another way, lets say a pixel is almost zero, or black. Using one pixel integers, it would round down to black, but averaging more than one pixel, one might find it wasn't quite black anymore. We have something between zero and one, i.e. greater dynamic range.

Except that's not what range is.

Re:Noise/Light Sensitivity/Optics

[Threni]

> Pixel size, per se, has no impact on the light sensitivity of the pixel.

Well, it does, because more light can fall on a larger sensor. Why do you think people pay loads extra for full-frame cameras? They're more sensitive, and less noisy, for a given iso/shutter speed/aperature combination, because the sampling is more accurate, especially in low light situations (noise isn't an issue in bright situations even for fairly cheesy cameras).

I'd imagine that a 120mp sensors is going to be as noisy as fuck.

Re:Noise/Light Sensitivity/Optics

[bigrockpeltr]

your argument hold for overexposure but (using your analogy) I'm sure that if you had a very very light drizzle having four containers you would at least catch one drop of rain. which is what he is referring to.

Re:Noise/Light Sensitivity/Optics

[Anonymous Coward]

Since you haven't explained what you think it is, your comment is worthless.

Re:Noise/Light Sensitivity/Optics

[Peeteriz]

In principle, you get the exact same result or worse as with a cheaper sensor with less resolution where each pixel is simply 4 times larger and gets 4 times the light for the dark areas, and the bright parts will be maxed out anyway. And HDR usually means a much larger exposure difference than simply 4 times - say, 10 stop difference is a 2^10 ~= 1000 times more light for the dark parts.

Re:Noise/Light Sensitivity/Optics

[spun]

Losing sharpness, otherwise known as trading resolution for dynamic range, is exactly what I was talking about so your argument that we would lose sharpness makes no sense.

In a church with no lighting, one pixel sensor may not pick up enough photons to activate at the lowest level, but several pixels combined, being a larger area, might pick up enough photons. I'm really unclear as to why this idea seems so controversial to some, but completely obvious to others.

Re:Noise/Light Sensitivity/Optics

[spun]

Lets say that on one gray scale sensor there are 254 integers between black (0) and white (255). On another one, there are (considering rounding errors) 1024 meaningful reals between black and white. The range is still 0 to 255, but we have more shades of gray between black and white, and if there is some other meaning to "greater dynamic range" than "more shades of gray" I certainly don't know what it is.

Re:Noise/Light Sensitivity/Optics

[Threni]

So... a 3x3 grid of pixels makes up 1 actual pixel, so it gets enough light. I'm missing the bit when you turn that 1 pixel back into 9 pixels. Haven't you just killed your resolution?

Re:Noise/Light Sensitivity/Optics

[Anonymous Coward]

But that's for 8-bit images. Most current DSLRs are capable of 14- or 16-bit RAW images.

Re:Noise/Light Sensitivity/Optics

[spun]

The logic holds no matter the number of bits, unless you are saying that this sensor, which is a more advanced version of the one in their high end professional camera, wouldn't have 14-16 bit RAW images.

Re:Noise/Light Sensitivity/Optics

[spun]

Yes. That's the point. With that many pixels, you could trade off resolution for dynamic range with the flip of a switch, and still have more resolution than you need for full HD or large format printing.

Re:Noise/Light Sensitivity/Optics

[spitzak]

The problem is you are confusing what is properly called "resolution" with "range".

A 2-bit system that can store the numbers 0, 1, 2, and 3 has a *range* of 3 and a *resolution* of 1.0.

A 16-bit system that can store 65535 fractions between 0 and 1 has a *range* of 1 and a resolution of 1/65536. It is higher *resolution*, not *range*.

Re:Noise/Light Sensitivity/Optics

[John Whitley]

I'd bet that you could use that many megapixels to seriously boost dynamic range by averaging several adjacent pixels into one.

Simply put: no. Software "averaging" may smooth out noise, but it will not add information that was not present in the first place. Missing dynamic range at the hardware is just not there to be recovered in software. In digital camera sensors, dynamic range is limited by saturation of the sensor's photosites. Once a photosite has collected enough photons, it registers maximum charge -- information about any further photons collected at that photosite during the exposure is lost. In fact, adding more photosites per unit area increases the per-photosite noise and chip areal overhead. Noise reduces dynamic range at the low end, and less charge capacity per photosite reduces dynamic range at the high end.

As another poster notes, you might change the effective exposure received by each photosite (perhaps by Bayer-array like neutral-density filtering). Or you can do what Fuji did with the S3 pro: make a matrix of photosites of different sizes/sensitivites to improve dynamic range. Fuji's sensor, while nice, has hardly taken over the digital imaging world.

On a more constructive note, Ctein wrote up a nice exposition on The Online Photographer about both near-term sensor technologies entering production and long-term avenues for advancement in digital imaging technology.

Re:Noise/Light Sensitivity/Optics

[spitzak]

It is a legitimate argument to say that averaging several pixels together can lower the noise floor, and thus lower the bottom edge of the range. If you then stop down the lens to take advantage of this you can translate this into raising the top end of the range.

However the result is tiny compared to using those pixels with filters above them so each covers it's own range.

Re:Noise/Light Sensitivity/Optics

[bws111]

There are no sensors that have 'reals'. There are (cheap) sensors that have 256 integers, and there are sensors that have 1024 (and more) integers. The sensors are basically counting the number of photons that hit them during the exposure. A sensor that has 1024 integers can capture 4 times the light as one that has 256. Therefore, the exposure can be 4 times longer without overflowing for the same lighting. During that 4 times longer exposure, some of the pixels that are on a darker area may capture some photons that they would not if the exposure was shorter. That is how the dynamic range gets extended. Your averaging method can not do that. If the exposure is short enough to not overflow the sensor, then some darker (but not black) areas will not have any photons registered, and will show up on the picture as black. If you average those dark areas you still get zero. If you make the exposure longer you can capture some shadow detail, but the bright areas will have overflowed and show as blown out. If you average those areas you still get the max number. The only thing averaging can do is insert a (false) value between adjacent dark and light areas, which improves neither the dynamic range nor the picture.

Re:Noise/Light Sensitivity/Optics

[dangitman]

and if there is some other meaning to "greater dynamic range" than "more shades of gray" I certainly don't know what it is.

Well, then you don't know the standard definition of "dynamic range" as it applies to photography. It means the difference between the highest and lowest light intensity a medium can capture. It does not mean "more shades of gray"

For example, a medium which can capture 1,000 shades of gray between EV-4 and EV-6 has low dynamic range compared to one which can capture only 100 shades of gray between EV-1 and EV-9.

Re:Noise/Light Sensitivity/Optics

[spun]

How is the dynamic range missing in the first place? Simple case, we have two black and white sensors next to each other. One is receives the minimum number of photons in a given period to register a "1," or the next darkest shade to absolute black. The one next to it registers a zero. Adding them together, we get .5, which is a shade that is in between 0 and 1.

Re:Noise/Light Sensitivity/Optics

[spun]

But averaging pixels would increase th4e actual dynamic range in the dark end. Consider two black and white sensors with a 0 to 255 range side by side. You have a cutoff which says, "x number of photons have to hit the sensor to register a 1" and you control the absolute amount of light in with the aperture. By averaging together adjacent pixels, you will pick up more photons for that one, larger pixel than you would have for the two smaller ones. Remember, this is happening in the camera, before any other processing. I think you could get a larger range, just as you could if you used a larger sensor. But I could be wrong, I'm not an expert in the field.

Re:Noise/Light Sensitivity/Optics

[spun]

Given that one pixel in a dark but not black area might register a one, and the one next to it a zero, averaging them gives you a .5, not zero. You aren't inserting a false point between two pixels, you are using what is effectively a larger pixel that can capture more photons.

Re:Noise/Light Sensitivity/Optics

[GlassHeart]

It's not possible to get more range out of a single exposure, because the range is inherent in the capture based on how much light you choose to let in, and how sensitive your sensor is to that light. Dynamic range refers to the difference between the brightest and the darkest pixel the sensor can distinguish in that exposure. Beyond the bright end of the range, they all look the same white to the sensor. Beyond the dark end of the range, they also all look the same black.

Here's how HDR works, oversimplified. We take a shot where we meter the bright part, so that it'll be properly exposed, deliberately sacrificing the dark parts. All dark pixels will be black in this exposure because we didn't let in enough light for the sensor to make out the difference. We then take another shot where we meter the dark part, sacrificing any somewhat bright parts. All bright pixels will be white because we let in too much light for the sensor to make out the difference. If we then combine the two images by throwing away the dark parts of the bright shot and the bright parts of the dark shot, we get an composited image that has more range than either image alone, i.e., HDR. Note that no averaging is involved.

The alternate solution ceoyoyo is talking about requires a different kind of sensor. Imagine if you had two kinds of pixel sensors, one sensitive and the other insensitive. You'd alternate them on your sensor, perhaps in a checkerboard pattern, but basically pairing adjacent sensitive/insensitive pixels. Now, if your sensitive pixel registers too high a value, then it's probably blown out so use the value from the insensitive one (which is by definition not as bright). If the insensitive one registers too low a value, then it's probably too dark, so use the sensitive one (by definition not as dark). The crucial difference here is that you choose one over the other, and never average. If all you did was average, then the result is the same as using a single kind of pixel sensor with a sensitivity in the middle, and would not improve your dynamic range.

Re:Noise/Light Sensitivity/Optics

[fyngyrz]

They don't even have a colour for each pixel yet

Actually, they do; see Sigma's foveon sensor for the SD-14, for instance.

Re:Noise/Light Sensitivity/Optics

[Threni]

Isn't that like zooming in optically, then zooming out digitally?

Re:Noise/Light Sensitivity/Optics

[scdeimos]

The process you're describing is called "binning" and is used extensively in astrophotography.

It should be mathematically equivalent, but you're almost always better using a lower resolution sensor because it has larger wells for photon gathering and is less affected by readout noise.

Binning doesn't improve the dynamic range of an image - that it restricted by the number of bits used in the ADC that the CMOS/CCD sensor pixels get clocked out to. Having a 2x2 bin (4 pixels) gives you four times the range of a single pixel (e.g.: 0-1023 instead of 0-255) but also gives you four times the sensor noise, so it's a no-win.

Re:Noise/Light Sensitivity/Optics

[Khashishi]

No, you wouldn't, unless it's your digitizer saturating instead of your CCD.
The maximum number of counts that a pixel can read is determined by the full well capacity of the CCD pixel. The full well capacity for a bigger pixel is considerably larger than that for a smaller pixel, so you don't get an advantage in dynamic range with smaller pixels. Actually, the opposite is usually true, since one big pixel holds more than 4 small ones.

Re:Noise/Light Sensitivity/Optics

[Khashishi]

I don't think anyone can make LCD pixels as small as CCD pixels.

Re:Noise/Light Sensitivity/Optics

[GlassHeart]

No, you are still confusing resolution and range.

In reality (on Earth), there's no such thing as infinitely dark. On a digital image, however, there's 0 which is the darkest possible value. Now, there a range of darkness below a certain point that a sensor cannot make out, and will register as all black (0). There's also a range of brightness beyond which the sensor cannot make out, and will register as all white (1.0, 255, whatever). This is the range of the sensor. How many discrete steps the sensor can differentiate between these two points is the resolution. More resolution does not mean the sensor can distinguish a darker point or a brighter point than it used to, so it doesn't actually increase the range.

Re:Noise/Light Sensitivity/Optics

[Estanislao Mart�nez]

I'm just curious what this would be like in low light settings, cramming that many pixels into such a small space has got to have some effect on sensitivity.

The effect of more pixels has led to a myth that more pixels = more noise. But this is not really the case. The low-light performance of a camera sensor really comes down to two things:

1. The size of the sensor. Larger sensor = more light = more signal = higher SNR.
2. The efficiency of the sensor--how good it is at extracting signal out of the light that hits it, and how "quiet" its own electronics are.

For a fixed sensor size and efficiency level, more pixels means that you get more noise per individual pixel--but the noise of adjacent pixels cancels out when you scale the image down.

Re:Noise/Light Sensitivity/Optics

[Anonymous Coward]

Well, you get exactly the same as you'd get with a sensor that has 4 times less resolution (CCD size kept constant, obviously).

Re:Noise/Light Sensitivity/Optics

[spun]

Is it? If you zoom in digitally, aren't you essentially cropping then enlarging the cropped area, throwing away pixels?

Re:Noise/Light Sensitivity/Optics

[spun]

Yes I get that and I guess I was presenting the argument in the wrong way, I'm saying (and I could be wrong) that I think that by essentially combining pixels and thereby giving a larger light gathering area per pixel, one could capture more photons per pixel. If the camera could treat the sensor that way (and maybe it can't?) wouldn't capturing more photons per pixel allow you to capture a larger range of brightness levels?

Re:Noise/Light Sensitivity/Optics

[martas]

what you describe has already been done, and ignore the people saying it's impossible. see http://www.cs.cmu.edu/~srinivas/

Re:Noise/Light Sensitivity/Optics

[bws111]

Let's say we have an image of 8 gray bars, with the brightest one 8 times as bright as the darkest. If we have a 3-bit sensor, with a resolution of 2 pixels per bar, we could adjust the exposure so the brightest bar had a value of 7, and we get the following pixel values:

0 0 1 1 2 2 3 3 4 4 5 5 6 6 7 7 and after averaging, we would have 0 1 2 3 4 5 6 7, which is an accurate representation. This sensor has enough dynamic range for the picture.

If we take the same picture with a 2-bit sensor, and again adjust the exposure so the brightest bar has the value 3, we have the following pixel values:

0 0 0 0 0 0 0 0 0 0 1 1 2 2 3 3 and after averaging, we would have 0 0 0 0 0 1 2 3. The entire left side of the image is black, all shadows are gone. Averaging did not fix that.

If we expose so we keep the shadows, we wind up with pixel values of 0 1 2 3 3 3 3 3 after averaging. The entire right side of the image is blown. Averaging did not fix that. If we expose for the center, we end up with averaged pixels of 0 0 0 0 1 2 3 3 3 3, with the shadows gone and the hilites blown.

Averaging can not make up for insufficient dynamic range.

Re:Noise/Light Sensitivity/Optics

[cheater512]

No you can do it in software, where some pixels have longer exposures than others.

There is your multiple exposures, however they would not be different as they were taken at the exact same time.

Some N900 software does the same thing but slightly differently, where it does a long exposure, but 'dumps' a short exposure picture from the sensor while its still capturing.

Re:Noise/Light Sensitivity/Optics

[Bill Dog]

If I understand correctly your comments so far, it looks like you're assuming that sensitivity is at least mostly a function of sensor pixel size, and therefore logically combining adjacent physical pixels would, only the the low end, extend the sensitivity range. I don't know that it works that way, as for all I know the most sensitive sensors require say x photons to hit them during an exposure to register anything, and say the bottom half of a scene is so dark that generally only x - 1 photons hit the pixels on the whole bottom half of the CCD. Then no amount of combining pixels down there will cause any of those pixels to register anything.

Re:Noise/Light Sensitivity/Optics

[Bill Dog]

(Bad edit at the last second on my above comment: I had it originally as "x - n", where I wanted it understood to be that n would always be at least 1. I.e. all pixels in the bottom-half area bombarded by varying numbers of photons, but all below the detection level of the pixels.)

Re:Noise/Light Sensitivity/Optics

[ceoyoyo]

I see what you're getting at, and it might be useful in very specialized circumstances, but probably not in a camera. Modern sensors usually only need a few photons to produce a signal. The problem is figuring out what is signal and what is noise. So your four little pixels are always putting out a signal, it's just too noisy to make anything out of.

You can certainly average adjacent pixels to reduce noise. The effect isn't linear though, you get a sqrt(N) noise reduction by averaging over N pixels. Because of the extra real estate you lose on your sensor due to support for each pixel, you're better off just having one big pixel instead of lots of little ones you average. But you're doing noise reduction, not increasing dynamic range.

Limits on dynamic range really aren't due to the amount of light falling on the sensor, but rather to the way the sensor is designed to accumulate charge. It's possible to make sensors that do capture a wide dynamic range, but they're expensive and the average person doesn't really use them anyway. Current cameras generally capture quite a bit more dynamic range than most people use. You'll even find a lot of people making "HDR" photos by taking a single RAW exposure and processing it.

Re:Noise/Light Sensitivity/Optics

[ceoyoyo]

Same as your eye does. Interesting, hey?

Re:Noise/Light Sensitivity/Optics

[ceoyoyo]

A couple of problems. A liquid crystal only changes the polarization of light. So to make it actually block light you need to polarize it. That involves a polarizing filter, which will always block half the light anyway. So now you've just thrown away half your light, across the board.

Second, I'm not sure a liquid crystal would be fast enough. Wikipedia says LCDs may not be able to actually switch fast enough to keep up with the latest 240 Hz LCD TVs. That suggests the maximum "shutter speed" you could get would be something south of 1/240th, which is pretty slow for most situations when you'd actually want to do HDR.

A better approach would be to make a sensor with an electronic shutter for each pixel. Most sensors have a global electronic shutter and many make use of it, including most compacts and most of the Nikon SLR line.

Of course, the best way (and quite likely cheaper than either of the other methods) would be just to make a natively high dynamic range sensor.

Re:Noise/Light Sensitivity/Optics

[ceoyoyo]

Since it takes a certain amount of chip area to support each pixel, breaking your sensor into more pixels and then averaging them back together, as the original poster suggests, LOSES you noise performance (and therefore dynamic range, according to your definition), rather than increasing it.

Re:Noise/Light Sensitivity/Optics

[ceoyoyo]

You could design a sensor that had adjacent pixels with different sensitivities. That's hardware. You can't do it in software, at least not usefully. Current sensors do not support individual pixel addressing or gain adjustment. At best you have to read a whole line at a time.

The trick of taking an intermediate reading off the sensor doesn't benefit from having extra pixels, or adjusting pixel sensitivities individually.

Re:Noise/Light Sensitivity/Optics

[danpbrowning]

As long as the rate of decline in full-well capacity matches the rate of increase in spatial-frequency-normalized read noise on the slope of pixel size, the dynamic range will stay the same, not decrease. In practice, The rate of decline is usually slower, particularly if you restrict the dynamic range to the circumstance of a typical photographer, who doesn't care for SNR lower than 8:1, and therefore will be photon-shot-noise dominated, leaving the FWC, not read noise, as the dominant factor.

That why dynamic range is so good in small pixels (if you factor out the overall sensor size, of course), like 10.3 stops in the cheap G10 at SNR 1:1, which slightly more then the $2,500 Canon 5D2 but not as good as the 12-stop D3X for $8,000.

The point is this: 120 MP allows for a choice between resolution and dynamic range. One the one hand, full resolution and normal dynamic range (e.g. 10 stops at 1:1 SNR), or less resolution and more dynamic range (e.g. 30 MP+11 stops, or 8 MP and 12 stops).

Unfortunately, this is not the case for low light performance, because it is dominated by read noise instead of photon shot noise. Unless a method is found to scale read noise linearly with pixel size, larger pixels will generally have an advantage over small pixels even after normalizing for spatial frequency.

Re:Noise/Light Sensitivity/Optics

[ceoyoyo]

Since we're talking about HDR, those changes in dynamic range are basically rounding errors.

Re:Noise/Light Sensitivity/Optics

[Anonymous Coward]

Talking about integers and reals is essentially adding precision when there is none. For example, suppose my I average 0, 1 and 1. I get 0.666666... but that doesn't mean my measurement has infinite precision. Ultimately that real has to go back to an integer and in this case, all you can say is the pixel value is 1. You might say you're reducing measurement noise but that's completely different than dynamic range.

The reality of dynamic range is that you have two sensors in different parts of the image. One area is dark, the other light (like the force.. duct tape, etc...) and high dynamic range means neither region is underexposed or overexposed. You need to make sure the dark area gets enough photons to measure something meaningful at the same time you have to make sure the light area is not saturated. Averaging 0 and 0 is still 0. Averaging 255 and 255 is still 255.

In your technique how can you make sure both the light and dark areas of the photo are not under or over exposed?

Re:Noise/Light Sensitivity/Optics

[Mudd Guy]

Actually, it is possible to get more DR (in fact enough for most HDR applications) out of a single sensor, but you need two amplifier/digitizer chains. Typical sensors in DSLR cameras have at least 2 stops of DR that is lost after the sensor! Here is a good discussion of this.

Re:Noise/Light Sensitivity/Optics

[Threni]

Who mentioned zooming in digitally?

Re:Noise/Light Sensitivity/Optics

[spun]

Oh, duh. Bad eyeballs, bad. Yeah, I guess it is like that.

Re:Noise/Light Sensitivity/Optics

[RealGrouchy]

The alternate solution ceoyoyo is talking about requires a different kind of sensor. Imagine if you had two kinds of pixel sensors, one sensitive and the other insensitive. You'd alternate them on your sensor, perhaps in a checkerboard pattern, but basically pairing adjacent sensitive/insensitive pixels. Now, if your sensitive pixel registers too high a value, then it's probably blown out so use the value from the insensitive one (which is by definition not as bright). If the insensitive one registers too low a value, then it's probably too dark, so use the sensitive one (by definition not as dark).

So wouldn't that screw with high-contrast detail images, resulting in something like this?

- RG>

Re:Noise/Light Sensitivity/Optics

[GlassHeart]

Yes, there are probably going to be odd artifacts to this solution. For one, you'll only be using half the light that hits your sensor, which probably means a bigger sensor and lens to compensate. The solution in Mudd Guy's link above is probably a more interesting path to pursue.

Re:Noise/Light Sensitivity/Optics

[holmstar]

Lets say the actual intensity of light varies from 0 to 100, and your sensor is capable of measuring values between 10 and 90. Below 10, the sensor considers the pixel completely black, and above 90, completely white.

if, say you were to have four pixels, and one of the four registers a value of 11, while the other three register a value of 10, then the average is 10.25. Likewise, on the other end of the spectrum, if 3 register 90, and one registers 89, the average is 89.75. So averaging didn't allow the sensor to read values above 90, or below 10

In photography, the dynamic range is the distance between the top and bottom of what the sensor can register. Your example doesn't extend the range. Instead, it just breaks the range up into smaller pieces.

Do you see what we are getting at now?

Re:Noise/Light Sensitivity/Optics

[holmstar]

correction, as not to confuse...

Lets say the actual intensity of light varies from 0 to 100, and your sensor is capable of measuring values between 10 and 90. At 10, the sensor considers the pixel completely black, and at 90, completely white.

Re:Noise/Light Sensitivity/Optics

[holmstar]

wouldn't capturing more photons per pixel allow you to capture a larger range of brightness levels?

no, because the sensitivity of each physical pixel sensor is the same. It is that sensitivity that defines the range that your overall sensor can provide. You could do some post processing effects like multiplying the values of adjacent pixels, but the dynamic range is still limited by the sensitivity of the pixels themselves.

Re:Noise/Light Sensitivity/Optics

[hazydave]

Most DSLRs can capture in RAW mode, which gives them the expected 14-bits out of the ADC. If there's more range to be had from the sensors themselves, you could certainly add a second ADC (well, in theory... CMOS chips already have tons of ADCs, rather than the single unit of the CCDs) with a different gain. But then again, you could just add more resolution to the one you've got.

Though there's a Tascam field recording doing this with audio... one full sensitivity ADC, one with an attenuator in front of it. If you're clipping on the first one, you'll have a usable recording on the second. In short, it's extended the dynamic range beyond the 24(-ish) bits of the ADCs involved.

The idea already done by Fujifilm is to build a sensor with two different sensor site sizes, so one's inherently more sensitive than the other. That would seem a more valuable use of the technology than making all of these tiny, tiny sensors.

Re:Noise/Light Sensitivity/Optics

[hazydave]

Most single-sensor video cameras are already using pixel binning these days. You'll find 6-12 Mpixel sensors in most of these, but the target is, of course, a 2Mpixel image. It actually makes sense with a video camera... with a single sensor and large pixels, you have Bayer color distortion problems... you're interpolating two out of three colors for each pixel site. Go to an 8Mpixel sensor, and you can bin four sensors per video pixel, no interpolation needed.

They're going to have to do very much the same thing to get usable video out of this thing.... they could presumably bin 8x8 pixel chunks for an HD output.

If you have truly random noise, though, you don't get four (or 64) times the noise. Random noise is just as likely to lower as raise a pixel value, so adding them together sums both positive and negative noise contributions. Pattern noise (predictable sensor noise) is already dealt with very well in modern sensors.

Re:Noise/Light Sensitivity/Optics

[hazydave]

I get a pixel size of 2.2um x 2.2um maximum... that's about what you'll get on an 8Mpixel, 1/3.2" sensor. So yeah, the low-light performance at 120Mpixel is going to be about that of a typical compact P&S or mid-range consumer camcorder.

The one saving grace there... the noise pattern is going to be vanishingly small compared to a camcorder or 8Mpixel camera. So you'll have the same level of noise, but as long as it's random noise, it won't be nearly as visible. This is the same reason as video resolutions increase, we can get away with higher compression levels. For example, ATSC television is 6x the spatial resolution of DVD, but only about 2.5x the bitrate of DVD (DVD is, of course, variable bitrate). Compression artifacts can still be there, but they're much less noticeable.

The real question I have is entirely different... where in the universe is anyone going to find a lens worthy of such a sensor. When you translate to usual resolution numbers, this sensor is 230lp/mm. A very, very good lens for a 35mm camera will have a "Lens MTF" of 60lp/mm-80lp/mm. This is the resolution at which point the lens, at its best setting, produces a 50% MTF rating (eg, the contrast of the line pair is reduced to 50% of ideal). So in order to actually get any real value out of this in a DSLR, they're going to need a lens 3x-4x sharper than the typical $2,000+ Canon L-Series lenses of today. Good luck with that....

Re:Noise/Light Sensitivity/Optics

[hazydave]

An LNA adds noise... all amplifiers do. Think about it -- you're amplifying both signal and noise by the same amount. A perfectly noiseless amplifier would do nothing to the S/N. But such critters don't exist.

In an RF system (I design these), you do calculations based on noise factor, which is a device's ability to add noise to the system. Some of this isn't obvious... for example, an attenuator can also add noise to a system.. push the signal down into the noise floor, and you've added noise just as effectively as any amplifier.

The idea behind the LNA is that you're going to boost up a weak signal as quietly as possible, so that noise factors further down the cascade won't have as much of an effect on the signal. In any system, the noise factor of the first amplifier dominates.

Of course, in a CMOS chip, you have an amplifier and an ADC very close to any given sensor. You need that amplifier to boost the signal to the point it can be digitized. In strong light, you're sampling well above the noise floor, so there's no problem. In weaker light, you may very well have to boost the sensor output enough that the noise floor becomes visible. There isn't likely to be a system of amplifiers, probably just the sensor amp itself plus the PGA that sits before the ADC. And once you're digitized, of course, noise is no longer being added. Well, at least until you do image processing or make a JPEG :-)

Re:Noise/Light Sensitivity/Optics

[hazydave]

Pixel size has a huge impact on the sensitivity of a pixel. The level you get out of that sensor is based on the number of photos that hit that sensor in any sampling period. If my sensor is 4x the area, on the average I'll have 4x as many pixels hitting the sensor -- thus, 4x the output level, all things being equal. Thus, to match that output, the smaller sensor will need 4x the amplification.

Now, in bright light, you'll never know the difference. But as it gets darker, that smaller sensor's going to be pushing noise up into visible levels 2 full stops before the larger sensor needs that level of amplification.

This is why single sensor DSLRs are producing dramatically better low light video than professional 2/3" video gear 10x-20x the price.

Re:Noise/Light Sensitivity/Optics

[hazydave]

Er... that's "photons", not "photos" or "pixels"... my editor has the day off.

Re:Noise/Light Sensitivity/Optics

[hazydave]

All a micro-lens can do for you is make a smaller sensor seem to occupy the full sensor-cell space available to it. So if you have a 2.2mm x 2.2mm sensor site (as in this 120MPixel example), the micro lens gives you the effect of a 2.2mm x 2.2mm sensor, even if in the actual silicon, it's smaller.

However, you can't steal light from your neighbors. Compare this to a Canon 5D Mk II sensor... the pixel pitch on this model is 6.4mm x 6.4mm. That's nearly 8.5x the area per pixel. There's absolutely nothing a microlens can do to make this any better for the 120Mpixel sensor.

Re:Noise/Light Sensitivity/Optics

[danpbrowning]

Compare this to a Canon 5D Mk II sensor... the pixel pitch on this model is 6.4mm x 6.4mm. That's nearly 8.5x the area per pixel.

Yes, a single 6.4 micron pixel has much higher sensitivity than a single 2.2 micron pixel, but that's not seeing the forest for the trees. The higher number of pixels perfectly compensates for the lower sensitivity in the same way that a eight small cylinders can achieve the same total horsepower as four large cylinders if the displacement is the same for both engines. It's not the horsepower-per-cylinder that matters, it's the total horsepower of the engine.

The reason for this equivalence as it pertains to sensitivity is that the noise power at any given spatial frequency (level of detail) is equal as long as the sensitivity per square micron is equal.

Smaller pixels sample detail at a higher spatial frequency. If the noise power at that higher level of detail is compared directly to larger pixels which sample lower frequencies, it will be concluded the small pixel sensor is worse overall when in fact it is not, just as measuring hp-per-cylinder would lead to the incorrect conclusion that a small engine with large cylinders is better than a large engine with small cylinders. In reality, the small pixel sensor achieves the same noise power as the large pixel sensor for any given spatial frequency, it just has the added and optional capability to sample higher spatial frequencies.

The 5D2 you brought up is an excellent illustration of this. Despite having much larger pixels than the 7D, its noise power for any given level of detail and amount of sensor area is the exact same.

(Since we're only talking about sensitivity and not read noise, all this of course applies only to circumstances where the system noise is dominated photon shot noise.)

All a micro-lens can do for you is make a smaller sensor seem to occupy the full sensor-cell space available to it.

Right, and that is precisely what allows it to have the same sensitivity (photoelectron carriers per square micron), whereas otherwise a larger portion of the pixel area would have been taken by non-photodiode elements (for a given minimum feature size). Without microlenses, not only would smaller pixels have lower sensitivity per pixel, but they would also have lower sensitivity per area (e-/um^2).

Re:Noise/Light Sensitivity/Optics

[hazydave]

And then there's the problem of diffraction. For a 35mm lens, the physical limit for an f2.8 lens, on green-yellow light, is 247lp/mm at MTF 50%, 104lp/mm at MTF 80%. To meet or beat that 230lp/mm at 80% MTF, you'll need about an f1.2 lens.

In short, no such better lens is going to exist for a 35mm camera, period. The only way to really take advantage of 120Mpixels is to go to medium format digital. Hasselblad already sells a 60Mpixel digital back.

Re:Noise/Light Sensitivity/Optics

[crgrace]

Dave,

I agree with you, the LNA adds noise (of course it does). My point was that the SNR of the system with the LNA is better than the SNR of the system without the LNA. The noise figure of most ADCs is atrocious, you know.

For the sensor, there are three things at work. First, the vast majority of CMOS sensors have source-follower amps in the pixel to drive the output bus. These have gains less than one, and increasing their gain as much as possible improves the dynamic range. Second, there is some "charge conversion gain" associated with the photodiode that is actually the sensor. There is a lot of work in optimizing this. Third, there is the PGA and ADC you mentioned. To lower the noise floor of the pixel to below that of the electronics, people generally work on the photodiode, since, in a sense, it is analogous to the LNA in an RF system.

Carl

Re:Noise/Light Sensitivity/Optics

[hazydave]

I sort of agree. Of course, a microlens is a convex circular/semisphere, depending on the particulars. So you only at best cover PI/4 of a unit sensor site.

But of course, you can't even have a microlens covering quite the full circle. In-between each sensor site, you have baffles, to prevent light scattering from one site to the next. These take up a finite space... more of the total area, as you grow the number of pixels. I'm sure it's a fairly small nit to pick, but hey, this is slash-dot :-)

It's also rather interesting how this sensor is pretty much suggesting the limit on 35mm (or smaller) camera sensors. Just based on physics (diffraction based on wavelength), you can only ever get a 50% MTF at around 250lp/mm and f2.8. For 80% MTF in the same resolution, you need about f1.2. That's for yellow-green light; you'll do a little better under bluer light, a little worse under red light. So at the majority of working apertures, you're probably not going to be able to tell between a photo shot with a 120Mpixel sensor and one scaled up from an 60Mpixel sensor on a 35mm-class camera... and that's with a very good lens.

Re:Noise/Light Sensitivity/Optics

[danpbrowning]

Thank you for the great post, Dave.

you can only ever get a 50% MTF at around 250lp/mm and f2.8. For 80% MTF in the same resolution, you need about f1.2.

Excellent point. Those are the same numbers I get.

As I'm sure you know, a lot of contrast can be restored with sharpening or deconvolution. 50% MTF and even as low as 20% MTF can be restored to pretty high contrast. But it only works well if noise is low enough.

To me, it would be ideal to have a pixel size that results in less than 15% MTF for typical f-numbers (say, f/5.6, which puts us at about the same pixel size Canon used here). That way we could finally get rid of the anti-alias filter and improve MTF noticeably at a lot of lower spatial frequencies without aliasing artifacts.

So at the majority of working apertures, you're probably not going to be able to tell between a photo shot with a 120Mpixel sensor and one scaled up from an 60Mpixel sensor on a 35mm-class camera... and that's with a very good lens.

Agreed. Not to mention the much more common case where resolution is diminished even more by missed focus, motion blur, lens aberration, poor collimation, etc.

Yet despite all these reasons why most circumstances wont benefit from such small pixels, I would still like to see it happen. I think the downsides are small for most common use cases (ample light, normal dynamic range). The benefit will be for those rare but useful times when the resolution does come in handy, such as making teleconverters obsolete.

One of the most commonly cited downsides is file size. But for me, RED ONE has already proven that this is a solved problem: with the right software and sufficient encoding power in the camera, we can keep the overall quality to file size ratio the same at any given pixel count. So a 120 MP heavily compressed to 10 MB will give the same quality as a 20 MP compressed to 10 MB. Of course, in that situation the higher pixel count would not offer a benefit -- that is not the intention -- but it would at least allow for the pixel count to not be a disadvantage. Then the same pixel count can be used for lots of purposes. Those who have the space for large files can use lower compression or none at all, while those would prefer to keep the same file size as lower pixel count sensors can do that too.

Another oft-discussed disadvantage is processing speed, such as for raw conversion, but I think this too can be scaled linearly with overall image quality, at least for powers of two scaling factors, as demonstrated by the alternate demosaic in Raw Photo Processor. Higher quality solutions may be possible in the same time if execution is sufficiently parallelized and Moore's Law stays ahead of pixel count increases.

Re:Noise/Light Sensitivity/Optics

[ZosX]

Yeah, but the fuji camera you are referring to drops down to 6mp for its high sensitivity mode and the results aren't all that spectacular even at 6mp if you ask me. Sure you get more range, but at a huge loss in resolution, and that lens isn't really remotely sharp, so its probably resolving less than 6mp anyways. I really wanted to like the fuji cameras, but their lens system needs some serious work, and something tells me that as these lenses start approaching 30x of zoom, that the manufacturers are compromising IQ for something that sounds better to the average consumer. The Panny Leica designed 18x (27-486mm) lens was probably the sharpest of all the superzooms. Its too bad that they moved up to 24x, because the sharpness is pretty awful now. The FZ100 reads like a dream on paper, but has far worse image quality than my FZ28 and FZ35. You just can't simply make a decent 27-600mm lens.

Re:Noise/Light Sensitivity/Optics

[ZosX]

What are you talking about? My camera is 12mp and it outputs 4000x3000px files. That is *drum roll* 12 million pixels! Shocking right? So actually, to interpret RBG from 3 sensors (each with a color filter), there must be....36MP? The raw files are simply just the output of each sensor....the algorithm that interprets this data varies between raw processing softwares, but the concept takes the pixels and interpolates them into color using a color matrix. Interestingly, there are like 2x as many "green" pixels than red or blue because that is the color that the eye dominantly sees, so I have no idea how it manages to generate rgb, but I do know that they resulting files are definitely 12MP.

Re:Noise/Light Sensitivity/Optics

[ZosX]

Yeah...chdk is what makes the cheap canon P&S cameras really shine. Still wish I hadn't dropped my A590IS off a wall. Those cameras became impossible to find when the successors removed the manual controls.

Need some sharper glass... or better physics

[BWJones]

Canon had better come up with some sharper lenses with a sensor like this. I shoot shoot with APS-H sensors on the Canon 1D and many of the lenses that Canon, Nikon and Sigma among others make are not nearly sharp enough to deal with many more pixels than are on say... the Canon 1Ds. Zeiss makes some sharp glass, but with the pixel density Canon is talking about with this new sensor, I'd worry about noise in low light conditions like those on my last embed on the USS Toledo (world's first embed in a strategic nuclear submarine). Any sort of low light, high ISO images will be truly challenging environments for such small pixel imaging sites.

It might be a great technology demonstrator or even a specific use CMOS chip for longer exposures, but I doubt it will have any applications in consumer or professional cameras unless some additional technology (or physics) comes into play.

Also, one would have to come up with some new strategies for moving all of that data around. As it is, on the latest Canon 1D Mk IV, they are pushing 16.1 MP around at about 10 fps. With this new sensor, just the readout would prevent this sensor from being used in any but the most specialized of applications.

Re:Need some sharper glass... or better physics

[jvillain]

Cannon makes some awesome lenses. You just can't buy them in the toy department at Best Buy. The problem with high density sensors is that the denser they get the higher the noise level becomes. I think that is one of the reasons that Cannon isn't tripping over them selves to ramp up the Megapixal count that fast.

Re:Need some sharper glass... or better physics

[$RANDOMLUSER]

I'm sure the folks over at SDSS (Sloan Digital Sky Survey) would be happy to make use of a sensor like this.

Re:Need some sharper glass... or better physics

[localman57]

Plus, sooner or later the general public is going to realize that megapixels aren't everything. A the output of a 6 megapixel Nikon D40 will amaze your non-photographer friends, while the 14Megapixel Samsung compact you just bought at walmart will most definately not.

Re:Need some sharper glass... or better physics

[BWJones]

Canon does makes some great glass and I shoot exclusively with Canon glass. However, Nikon, Zeiss and Leica among others also produce some pretty sweet lenses. Eventually, everybody is going to have to deal with issues related to the optics being able to actually resolve the imaging sites. At some point (and we are close), the glass will not be able to resolve anything more than the sensor can read out and you'd have wasted pixels. Kinda like the issue with Apple's Retina Display on the iPhone 4 that I wrote about here. Any more pixels would be wasted given the resolving power of the human eye.

Re:Need some sharper glass... or better physics

[Anonymous Coward]

The problem with modern digital cameras is that they are diffraction limited, http://en.wikipedia.org/wiki/Diffraction-limited_system, by the laws of physics, or very nearly so, given current day lens technology. There is no way you will get a higher actual resolution without going to lenses, which are significantly larger in diameter than what we are used to in dSLRs. So adding more pixels in the area of the sensor of the latest Canon 1D models is completely pointless, which is why we haven't seen an update yet featuring higher resolution.

In other words: Keep dreaming. These new detectors are just marketing gimmicks, or intended for specialist scientific applications, like astronomy.

Re:Need some sharper glass... or better physics

[Anonymous Coward]

Canon Unveils 120-Megapixel Camera Sensor

Yeah that's great but the consumer market is waaay to caught up in megapixels this and that. When I bought my last digital body I did some extensive reading over at DPReview http://www.dpreview.com/reviews/ and I found little difference between two verions of the same model. Sure they're cramming in HD video and face recognition etc but if you can't make significant improvements in quality what's the point (ok sell camera's to uneducated consumers I know)? Case in point Canon xti and xsi, the reviewer found that the older model (xti) even preformed better in some situations than the newer (xsi). The newer model has more pixels and some nice bells and whistles but can you take a substantially better picture to justify the price. I think not. In fact as I was shopping I found a lot of places where they were the same price. That whole industry needs to clean up their act. Sorry for the diatribe but stuff like that really burns me up. Shoehornjob

Re:Need some sharper glass... or better physics

[Entropius]

Canon does make some awesome lenses, but even some of their L-lenses look somewhat lacking when used on their high-resolution sensors.

My father has a 70-200 f/2.8L. It actually shows pretty low contrast and a "hazy" look until you stop it down to f/4 or more, especially at the long end. The new 70-200 mk2 is much better.

Re:Need some sharper glass... or better physics

[Entropius]

Random question: My father shoots Canon, and has gotten sort of frustrated with the ADHD problem of the autofocus. Using two different lenses (70-200/2.8, 100-400) and two different bodies (350D, 500D), he's noticed that the AF is easily distracted by foreground clutter, and will also inexplicably refuse to confirm an AF lock (and thus shoot) in some situations you'd think are easy, like a bird on the end of a twig with a background distant enough to be a blur. Have you experienced anything like this? (This is in single-shot center-point AF mode.)

I have an old Olympus SLR, and if you can see even a tiny piece of a bird visible through the foreground brush, it'll lock right in and shoot, perfectly in focus.

Olympus doesn't make the lens I want (a 300 f/4), so I've thought about switching to Canon. But the AF scares me a bit.

Re:Need some sharper glass... or better physics

[Amouth]

i want this

http://graphics.stanford.edu/papers/lfcamera/ /. posted it here ~5 years ago.. still waiting

http://slashdot.org/articles/05/11/21/2316216.shtml?tid=126&tid=152

Re:Need some sharper glass... or better physics

[Anonymous Coward]

I would say to get an EF lens mount for your Olympus SLR and buy the lens you want, but this will only work if your SLR electronically controls the aperture AND your lens mount translates the signals correctly, and I don't know enough to say if this combination even exists. I lean towards guessing no.

Or you can just go with a mount if you don't mind having the aperture wide open. Or find a similar lens from another company.

Re:Need some sharper glass... or better physics

[lwsimon]

I agree. I use a D70 for my hobby photography, and I would feel comfortable with it even today in many professional settings. That camera is 7 years old now.

Re:Need some sharper glass... or better physics

[shadowfaxcrx]

It's not like that's unique to the photography industry. They're pushing terabyte drives now. I defy the normal computer user (hint: if you're reading this, you are not a normal computer user) to fill a terabyte. Ain't gonna happen.

Back in the 486/pentium days several makers were shipping machines with 2 cd-rom drives. Why? I still don't know. But 2 is better than 1!

The Bugatti Veyron has 1000 horsepower, and can do 253mph while using over a gallon of fuel per minute. No one who buys that car will ever be able to get anywhere close to its performance envelope without getting arrested or killed, but they're still selling it.

CCD specs are the photographic version of penis measurement. More is better, even if it makes no difference to the outcome of the activity.

Re:Need some sharper glass... or better physics

[RemyBR]

Your father's lens is probably in need of calibration. I use one of those and it shows none of this even when wide open. Or there's a change he got a bad copy, in which case calibration would still help, but not much.

Re:Need some sharper glass... or better physics

[T-Bone-T]

I've not experienced this but some of the bigger telephoto lenses have the option to ignore foreground objects. If you focus on something, it won't try to refocus on something much closer. If I was your dad, I'd turn off autofocus to get a good focus and then turn it back on.

Re:Need some sharper glass... or better physics

[Ichijo]

I shoot shoot with APS-H sensors on the Canon 1D and many of the lenses that Canon, Nikon and Sigma among others make are not nearly sharp enough to deal with many more pixels than are on say... the Canon 1Ds.

If you always shoot wide open, I can see why you would say that. But if I stop down, even the kit lens on my 10MP Canon XTi outresolves the sensor.

Re:Need some sharper glass... or better physics

[kimvette]

Considering that GP is discussing the EOS xD line, Zeiss lenses, and so on, I really doubt he's shopping for lenses at Best Buy.

However despite what you're saying, there are some hidden gems in Canon's cheaper lens offerings. I bought an EF-S 18-55mm f/3.5-5.6 IS may exhibit a bit of CA (easily corrected in post) as a throwaway lens (one just to use for an event and then replace it with better glass later) but I was surprised to find that is is wonderfully sharp and there are folks claim it compares favorably to wider L-series zoom lenses for contrast and sharpness. Likewise, the EF 70-300mm f/4-5.6 IS USM is often referred to as a "hidden L" lens (I like mine but I wish I had an EF-70-200mm f/2.8 IS USM + 2x teleconverter instead). There are some decent lens options you can find even at Best Buy, if you do your research. No, you won't get a weather-sealed ruggedized, constant f/2.8 or f/1.2 lens at Best Buy, but if you can get by with a slow lens that isn't weatherproof or ruggedized you can find lenses that are perfectly suited for your child's birthday parties and soccer games.

Canon makes optically wonderful lenses in both their L series and lower end lines, but the L-series lens isn't just for optical quality, but use where you need a ruggedized lens, and many are even weather sealed. Likewise they make quite a few cheaper zooms and primes that are wonderfully sharp. They may be slower (higher f-number/smaller aperture) but may boast sharp focus with great contrast.

But GP is right. Even the current 7D's pixel density (18MP on an APS-C size sensor) out-resolves a lot of older lens models. Try putting the 100-300mm USM on the 7D sometime; you won't get a sharp image. The 70-300mm IS USM, known for being extremely good optically, still looks a little soft on such a high pixel density. Canon has to start thinking about improving their lens offerings before they consider pushing the megapixel envelope any more. Besides, once you get to 12MP, for most uses, how is higher resolution a benefit? All you're going to do is eat up storage and spend more time downscaling images, and post will take longer.

Pixel densities of Canon's current pro DSLR offerings (source: dpreview):

EOS 1D mk IV APS-H 3.1 MP/cm pixel density 16.1 MP
EOS 1Ds mk III FF 2.4 MP/cm pixel density 21.1 MP
EOS 5D mk II FF 2.4 MP/cm pixel density 21 MP
EOS 7D APS-C 5.4 MP/cm pixel density 18 MP
EOS 50D APS-C 4.5 MP/cm pixel density 15.1 MP

With the higher pixel densities, focus sharpness and image stabilization become far more important. Imagine the effect of camera shake on a 120MP sensor? How about a lens that an 18MP APS-C outresolves? 18MP and 21MP are already enough - at least for me. I'd like to see sharper lenses please!

Frigging slashdot: "Filter error: Please use fewer 'junk' characters." Sorry about the lack of formatting on the table I typed up. With characters separating the columns slashdot was bitching about "junk characters" so I had to undo all the time I spent formatting the table. Stupid slashcode.

Re:Need some sharper glass... or better physics

[Entropius]

It's actually something that came up in DPReview's tests: see http://www.dpreview.com/lensreviews/canon_70-200_2p8_is_usm_c16/page4.asp . (Note that this is the IS version of the lens; the non-IS version has a less-sophisticated optical formula, I believe).

Re:Need some sharper glass... or better physics

[EastCoastSurfer]

Yes, pair that up with a 'nifty 50mm' and you've just blown away nearly every point and shoot out there. If you don't mind manual focusing (the D40 doesn't have it's own focus motor) then you can get the 50mm lens for $200. The AF one iirc, is right around $400 or less.

Re:Need some sharper glass... or better physics

[kimvette]

), he's noticed that the AF is easily distracted by foreground clutter, and will also inexplicably refuse to confirm an AF lock (and thus shoot) in some situations you'd think are easy, like a bird on the end of a twig with a background distant enough to be a blur.

It's a mix of two issues: first is focus point size. The second issue is that AF algorithms generally select the closest subject to the AF sensor. It sounds like you need a body that has spot AF (basically, it's a feature that cuts an AF point down to 1/3 its size) - or learn old-school focus and recompose. Why does AF scare you? Learn the features. Right now Canon's AF in the 7D is best of breed. All 19 focus points are cross-type phase sensors, and each AF point can be switched to "point" mode (shrink the AF point down to ~1/3 its size) and it works very, very well. The "point AF" mode will often let you focus on a subject between two blades of grass, where the normal-size single point AF mode would focus on the grass (closer object) instead. The 500D has only nine points, and the center point is rather large.

Why did I mention the 7D when you were referring to the consumer-level bodies? Because you are referring to L-series glass (with one of them, the 70-200 f.2.8 being weather sealed, being a perfect match for the weather-sealed 7D), and because of the advanced AF. If you want the best AF out there and you already have access to excellent telephoto lens, you'll want a body that will get you the most out of it. The AF shouldn't scare you.

Re:Need some sharper glass... or better physics

[jeffmeden]

I call BS, he must be doing something else wrong; the camera is dirty or not configured right, the lenses aren't genuine, he's just too close to the subject, etc. With my 400D and an array of cheap (but genuine Canon) lenses, center point one-shot AF on a centered subject is super fast and accurate in all but the lowest of lighting. Hell, I almost always leave my camera in AI-focus (notoriously picky) and it rarely lets me down.

Re:Need some sharper glass... or better physics

[jvillain]

My first solution to random autofocus issues is to shoot every thing twice. Shoot once then move the camera just a fraction and shoot again. Bracketing can help with that. Storage is cheap so shoot every thing twice if you can.

The second solution is that most Cannon cameras allow you to change the number and positions of the sensor points. The fully automatic modes usually wont let you but the semi auto ones will. The standard layout is roughly 10 dots covering the middle of the screen. But you can probably switch any of those one dots as the soul source for the focus. If you use the center one then lay that on the thing you want to be in focus you will be gold even shooting out of a bush. Consult the manual for info.

The third option is to go manual for the really tough shots. It can be faster to switch to manual that to press buttons getting the right dot selected. You can also some times find some interesting shots in manual that you wouldn't get in AF like moving the field of focus slightly do de-emphasise some thing and emphasise some thing else. There should be a switch on the lens to turn AF on and off on Cannon lenses.

Hope that helps.

Re:Need some sharper glass... or better physics

[danpbrowning]

It's true that some lenses are already into diminishing returns, particularly fast or wide lenses. But many other lenses are not even oversampled with 2 micron pixel sizes, including many macro and sharp primes. Take a look at this example of a $400 macro lens using 1.2 micron (simulated) pixel size:

http://forums.dpreview.com/forums/read.asp?forum=1019&message=29826265

Re:Need some sharper glass... or better physics

[danpbrowning]

Instead of worrying about "wasted pixels", we should be worrying about "wasted glass". A simple prime lens has about the same resolution and cost today as it did 20 years ago, whereas image sensor resolution and cost have advanced by many orders of magnitude in the same amount of time. Therefore we should be more concerned about that part of the system which is expensive and difficult to improve; not the part that is getting cheaper and better.

Re:Need some sharper glass... or better physics

[sco08y]

Plus, sooner or later the general public is going to realize that megapixels aren't everything.

Last I checked, unless it's an Apple product, any electronic item marketed to the "general public" is still festooned with idiotic stickers touting pointless numbers and techno-gibberish. (And even Apple does it, they're just a little more tasteful.)

I suspect it's not changing because current behavior is economically rational.

Purchasing an electronic device like a camera that's at the $200 price point. Say a poor purchasing decision will cost you maybe 20% of the price, so, typical loss form being uninformed is $40.

I suspect the effort to become very well informed about these things can easily dwarf $40. And, even if you are informed you can still buy a lemon.

Re:Need some sharper glass... or better physics

[pz]

With this new sensor, just the readout would prevent this sensor from being used in any but the most specialized of applications.

I'm thinking that you've hit the nail on the head: this is for specialized applications where things like light level are controlled and long read-out time can be tolerated. I'm familiar with some projects that are attempting to build nano-meter scale 3D reconstructions of the brain (they constructed the first gigapixel camera, although didn't announce it), and they would hugely benefit from a huge sensor like this.

I'd put good money on scientific and industrial applications.
 

Re:Need some sharper glass... or better physics

[Guido von Guido]

Cannon makes some awesome lenses. You just can't buy them in the toy department at Best Buy. The problem with high density sensors is that the denser they get the higher the noise level becomes. I think that is one of the reasons that Cannon isn't tripping over them selves to ramp up the Megapixal count that fast.

Although I've never actually seen them on display in a store, you can buy both Canon's high end cameras (like the 1Ds the grandparent shoots with) and their top-grade L lenses on Best Buy's web site.Not that I would recommend buying them from Best Buy...

Re:Need some sharper glass... or better physics

[Entropius]

Thanks for the advice. I do have a nice telephoto for Olympus, but I bought it used, and can't afford a top-end body. My dad can however, and I'll advise him to check out the 7D. Good to know that its focus points are just plain smaller than the 500D's. Still, that's not the only issue with the 500D (and 350D) autofocus -- sometimes they will just plain refuse to lock/shoot when there's no reason for it (isolated bird on twig seems to prompt this behavior).

The E-510 I have (old consumer Olympus body) has a fairly large center point, BUT if there is sufficient contrast at the very center it will focus on whatever's right under the dot. It seems to be the best of both worlds; the single-shot AF seems absolutely rock-solid. The one time I had trouble with it it turns out a bit of grass had gotten stuck right over the AF sensor -- remove it and it's great again.

Re:Need some sharper glass... or better physics

[Entropius]

The funny thing is, we've called Canon, and they say nothing is wrong. The 100-400L, at least, was bought new from B the 500D was also bought from B&H. I've checked the AF calibration using one of those calibration charts and it seems fine.

I'm a little befuddled too, since my ancient E-510 with a cheap lens will lock right in on some subject that he can't focus on, with any amount of monkeying with the AF.

Re:Need some sharper glass... or better physics

[Entropius]

He's already using center spot-focus. (That's what I leave my Olympus on too -- why should I let it pick what to focus on? I want to do that!)

Actually, most Canon lenses (and, for that matter, most Olympus ones too) have full-time manual autofocus -- you don't have to switch anything, just grab the ring and turn it. That's what he's resorted to doing, but it's harder without a split-prism. I dunno why they stopped putting those on viewfinders; you can get a third-party one for $100 for Olympus at least.

Re:Need some sharper glass... or better physics

[Anonymous Coward]

In what universe is the Los Angeles class attack sub USS Toledo a "strategic" nuclear submarine? Unless it has a few dozen nuclear warheads on top of some SLBMs, it's not "strategic" anything.

Re:Need some sharper glass... or better physics

[fyngyrz]

They're pushing terabyte drives now. I defy the normal computer user (hint: if you're reading this, you are not a normal computer user) to fill a terabyte.

Anyone with a video camera can fill a terabyte easily. And that would be almost everyone with a cellphone, these days, not to mention the millions of dedicated video recorders that are out there.

Re:Need some sharper glass... or better physics

[IICV]

... I've got fcamera on my N900, it works even better now that I've overclocked the phone a touch.

It was just a simple apt-get install fcamera away.

Re:Need some sharper glass... or better physics

[Jon Abbott]

I've been looking at ISO 12223 charts for Canon lenses for a while (that are published on The Digital Picture), and Canon certainly seems to be getting sharper with their newer lenses. The new TS-E 24mm f/3.5L II and the 70-200 f/2.8L IS II USM appear to be extremely sharp wide open. Don't know if you've had a chance to play with those but they look amazing. I agree that as megapixel counts go through the roof, we're going to start hitting diffraction limits that render the extra pixels worthless (more here for those who are interested).

P.S. Over the years I've seen many of your photos of the Bonneville Speed Week events, and really enjoy them. I've been meaning to get out there sometime to get some photos of my own. :^)

Re:Need some sharper glass... or better physics

[luder]

I own a 40D and an EF 50mm f/1.8 II lens. Can't say I ever experienced what you mention, specially when using the extra sensitive center-point. As long as the selected AF point is on the target, it usually focus accurately.

The only situation where that happens, in my experience, is when using all of the AF points, because it always focus on the closest subject. I find it mostly useless, so it's normally set in the center point. When shooting static stuff at close range using big apertures, I do select the closest AF point, since locking focus and recomposing can change the point of focus.

Other than that, maybe the bird is too small and doesn't "fill" the center-point, thus confusing the AF?

Re:Need some sharper glass... or better physics

[rcw-home]

My father shoots Canon, and has gotten sort of frustrated with the ADHD problem of the autofocus. Using two different lenses (70-200/2.8, 100-400) and two different bodies (350D, 500D), he's noticed that the AF is easily distracted by foreground clutter, and will also inexplicably refuse to confirm an AF lock (and thus shoot) in some situations you'd think are easy, like a bird on the end of a twig with a background distant enough to be a blur. Have you experienced anything like this? (This is in single-shot center-point AF mode.)

I have a 40D that I also mostly use in center-point AF mode. I mostly use a Tamron superzoom on it which at full telephoto is f/6.3, slightly out of Canon's autofocus spec (f/5.6). If the focus is way off to start with (like say you're focused up close and now you want to focus on that little bird 20 meters out) then the camera won't even know what you're pointing at (and you won't either if you're looking through the viewfinder). The circles of confusion from that stuff in the foreground probably overlap the bird entirely (maybe not at the aperture you're exposing at, but phase AF always works at the len's maximum aperture size). In that circumstance you'd need to get the focus reasonably close to the desired distance before it can lock on through the foreground clutter, either by letting it hunt or by locking on to something else at roughly the same distance without other stuff in the way.

In low-light situations (where I would be using 1/10 second or longer exposures at ISO 3200 without the flash AF assist) it starts to get very picky about what it will focus on, and at that point, if appropriate, I'll tell it to use all AF points again (this significantly improves sensitivity).

I also have a 50mm f/1.4 lens for it. One time while backpacking I decided to shoot a time lapse movie of the stars. I was able to reliably autofocus it on Venus before switching it to manual mode and turning on the timer remote.

I have also hooked it up to a Celestron C5 telescope, which is f/10, far beyond the f/5.6 spec. While the Celestron is manual focus only, I'm hacking some electronics gutted from a used 50mm f/1.8 onto it to make it autofocus. I'm not done yet, but I can easily coax a focus confirmation light out of it with modest indoor lighting.

It's the only SLR I've ever owned so I can't compare it to anything else. I hope what I've written helps.

Re:Need some sharper glass... or better physics

[shadowfaxcrx]

Cell phone, no, unless they're making about 50 trips a day between whatever they're shooting and the computer. A terabyte is about 68.25 hours of DVCPRO video (which is higher rez than most people shoot with, but even being generous and assuming you're shooting 1080 at 60i/30p, you still have 17 hours of video). Anyone shooting that much video is not shooting it with a cell phone, and is not a normal computer user.

Re:Need some sharper glass... or better physics

[viking80]

There is no problem to get a lens with better resolution than this. A fixed focus would certainly do just fine. At 750nm, and with n.a. ~1, a lens would need a diameter equal to the resolution. The diagonal resolution is sqrt(x^2+y^2) or 16146. So a stepped down lens should have d>12mm.

Secondly, if you look at a distant object, like a cell tower on a distant mountain, your eyes will resolve down to 20 seconds of arc (Sparrow's resolution limit). Your total field of view is 120x60 degrees (of couse not high resolution away from focus, but just to be able to make an image with same resolution as your eyes, and allow the eyes to wander the image), you will need: 120deg/20"=20kpix x 10kpix or 200Mpix.

I can not see how this is a lot, nor can I see how this can be hard for a simple lens to focus properly. I can see how a 10x optical zoom would be problematic, but well understood optics combined with a powerful DSP would fix that. Look at how the fuzzy images on Hubble (before the optics was fixed) was deconvolved with simple math.

Re:Need some sharper glass... or better physics

[BWJones]

Both the 70-200 and the 100-400 have focal plane switches so that if you are trying to restrict the focal plane to far objects only or include the entire focal range. The other trick you might want to do is use a single focus point. The 350D and 500D may not have the fastest focus processing, but there should be no problem using those lenses.

I gotta say that I've had no problems with autofocus on Canon cameras going back to the 20d I use to use.

Re:Need some sharper glass... or better physics

[BWJones]

Thanks. I've got one of Canon's new lenses in the 100mm macro f/2.8 L and it is wicked sharp.

I missed Bonneville this year, but am planning on going to World of Speed in September. Hope to see you there. If you go, look me up beforehand.

Re:Need some sharper glass... or better physics

[BWJones]

Among other interesting remote sensing applications, I am sure.

Re:Need some sharper glass... or better physics

[Moghedien]

You guess right: No adapter exists which will control a Canon EF lens on another SLR system. Not only is the aperture electronically controlled, some lenses have focus-by-wire, and cannot even be focused on another system.

Re:Need some sharper glass... or better physics

[Jon Abbott]

That weekend I'll be hosting some folks for the Great American Beer Fest... Perhaps next year I'll make it out there!

Re:Need some sharper glass... or better physics

[djdanlib]

Indeed. Some of my non-inclined friends are starting to notice this phenomenon as they upgrade their cell phones to new ones containing these horrendous 8+ megapixel cameras. Of course, as soon as I start talking about why that is, their eyes gloss over and I have to stop talking science...

Re:Need some sharper glass... or better physics

[djdanlib]

It's true - AF is a very selective creature and sometimes has a mind of its own. So, you have to learn how to work around its little quirks. When using a tripod, I would: aim the center-point at something exactly the same distance away like maybe another part of the subject, switch to AF, hold the trigger halfway down to set up the shot, switch to MF, confirm the focal point with the preview mode, re-point the camera and shoot. The order of this is important so you can go back to the beginning without any extra steps.

Don't forget about the depth of your focal plane, and how it increases with your aperture F# (as you make the aperture smaller). Sometimes this can save a picture.

Learn to use manual everything, and you will be able to creatively solve problems in the field. Age means nothing - I have worked with photographers well past the age of retirement who are better at this than I could have imagined.

Re:Need some sharper glass... or better physics

[fyngyrz]

mm, no, I don't buy it. An inexpensive HD video camera is $100; They're all over the place. Methinks you're figuring things using preconceptions picked up years ago.

As for "normal" hard drive use, I think the people who sell Aperture, Lightroom, Photoshop, WinImages, Finalcut, and all the plug-in people who write functions to extend those products... yeah, I think they'd just laugh at you. I know at least one of 'em is. :)

Re:Need some sharper glass... or better physics

[shadowfaxcrx]

So you're seriously suggesting that someone who fills a terabyte hard drive with raw HD video is a normal computer user? Or that someone who uses a pro-grade NLE, and a graphic editor that costs $700 is a normal computer user?

If you seriously think that then you're the one who should be laughed at. A *normal* computer user surfs the web, sends email, tweets everything about his life, and plays Solitaire. I'll even be generous and lump hardcore gamers into the normal category. But someone sitting in an edit bay using over 2 grand worth of software is not a normal computer user.

Re:Need some sharper glass... or better physics

[holmstar]

Back in the 486/pentium days several makers were shipping machines with 2 cd-rom drives. Why? I still don't know. But 2 is better than 1!

Less CD swapping, and the ability to run two programs that both required a CD at the same time.

The Bugatti Veyron has 1000 horsepower, and can do 253mph while using over a gallon of fuel per minute. No one who buys that car will ever be able to get anywhere close to its performance envelope without getting arrested or killed, but they're still selling it.

Because there is a market for it. If you are uber-rich, and find fast cars thrilling, what would be more thrilling than the fastest production car in the world? No, short of renting out a five-mile airstrip, or test track, you would never actually take the car to it's top speed. But the engineering required to allow that also bring benefits to the cars handling and stability at lower speeds, not to forget the obvious fun factor of having tons of torque.

CCD specs are the photographic version of penis measurement. More is better, even if it makes no difference to the outcome of the activity.

That you can't see the advantages of something does not mean that there are no advantages.

Re:Need some sharper glass... or better physics

[shadowfaxcrx]

Less CD swapping, and the ability to run two programs that both required a CD at the same time.

So. . you think people were playing 7th Guest and Myst at the same time?

Because there is a market for it. If you are uber-rich, and find fast cars thrilling, what would be more thrilling than the fastest production car in the world? No, short of renting out a five-mile airstrip, or test track, you would never actually take the car to it's top speed. But the engineering required to allow that also bring benefits to the cars handling and stability at lower speeds, not to forget the obvious fun factor of having tons of torque.

The market argument is an attempt to make it look like you're educating me, when really you're saying exactly what I'm saying. The point that myself and Anon were making is that there is a market for useless things, and the fact that there is one, is stupid.

That you can't see the advantages of something does not mean that there are no advantages.

Alright. Educate me. What is the advantage to having a >5mp point and shoot camera?

What is the advantage to having a 120mp camera at all, considering that's sharper than the lens can focus to?

Re:Need some sharper glass... or better physics

[hazydave]

That's actually what we're going to see with a 120Mpixel sensor. This one delivers 230lp/mm. You can get a theoretical maximum of about 250lp/mm on any 35mm lens at f2.8 with a 50% MTF. You can get all the way to 80% MTF at 230lp/mm going up to f1.2... but who's got an f1.2 lens. Ok, I used to have one too, for my old Canon rangefinder, but in multiple focal lengths? Stop down to f4.0 or so, and you won't be able to tell between the 120Mpixel sensor and a 60Mpixel sensor. And that's just due to diffraction limits.

Not that I'd have a problem trading in either of my DSLRs (Nikon D70, Canon EOS 350D) for one of these, mind you ;-)

Re:Need some sharper glass... or better physics

[hazydave]

Which Olympus (I still own an OM-1 and an OM-4)? If it's an automatic model, do you ALWAYS use auto-exposure? Olympus OM-series didn't do autofocus (well, there was the "failed experiment" of the OM-77

Auto-focus is no different.. there's a switch on any Canon lens to turn it off and let you focus manually. Most models also have a button that lets you move the auto-focus zone around, which is also useful. You may want a different ground glass screen if you do lots of manual focusing, though (and you get a Canon model that offers this). Due to the AF, most Canons (and other modern DSLRs) don't offer the hand-dandy focusing aids on the ground glass by default anymore.

As for the lens you want, there is an OM System Zuiko that's 300mm f4.5, about $300 these days on eBay. They also a 350mm at f2.8 -- still really expensive, if you can locate one.

Re:Need some sharper glass... or better physics

[hazydave]

Yup. I can fill a terabyte per project when I do multicamera shoots, easily. And do -- cheap SATA drives and a plug-in drive bay is a great way to move from project to project. I've got unedited Cineform files around here over 100GB. Ok, not "normal".. but also not what the average /. geek is likely to be doing, either. The average consumer is going to fill a drive faster than the average coder, unless the latter has a need to keep every ISO for every Linux distro at one's fingertips.

Re:Need some sharper glass... or better physics

[fyngyrz]

So you're seriously suggesting that someone who fills a terabyte hard drive with raw HD video is a normal computer user?

Yes, of course. Plug the USB cable in, drag the icon for your HD video... keep doing that, you'll fill the drive. Why take all those videos? Apparently you've never been a parent (but that's ok, most people eventually are parents.) So yeah, normal computer users.

Or that someone who uses a pro-grade NLE, and a graphic editor that costs $700 is a normal computer user?

Sometimes they are. Sometimes they aren't. But that wasn't the issue at hand; the issue at hand was, will normal users use up a terabyte, and the answer to that is still yes.

But someone sitting in an edit bay using over 2 grand worth of software is not a normal computer user.

You need to learn to parse English. No one has suggested any such thing here; where you got it, I don't know. But, since you seem to be ignorant of the details, of the software I mentioned, some of it is $50 (new) and some of it is more; all of it enables working with very large data files, some of it makes them out of thin air (no input video required.) People own those programs for various reasons, most of them *not* because they are "pro" users. And again, they're quite capable of chewing up a terabyte without breaking a sweat.

Heck, I've got a terabyte used here and there's almost nothing on that drive but JPEGs and RAWs from my consumer DSLR, a Canon.

Well, look. You're wrong, and every day that passes, you're just going to get more wrong, so by all means, have the last word. Perhaps it'll stand in for the feeling you'd get if you were actually correct. :)

Re:Need some sharper glass... or better physics

[shadowfaxcrx]

You talk about me having the last word, but you wander in an entire week after my last reply to blather on about how Final Cut editors are normal computer users. Up until your last sentence, I thought I was talking to someone who was simply misinformed. Now I realize that you're vapidly clueless.

You keep claiming that people who set up a multimedia editing suite are normal computer users. No one's going to buy your peculiar little brand of idiocy no matter how many veiled insults you add to the mix.

You might have convinced a few people if you hadn't thrown a pro-grade non-linear video editor in there just to show that you know the names of a few media applications, but using that program pretty much sunk your argument beyond salvageability - not that it wasn't already approaching a solid wall of stupidity when you tried to convince us in a spectacularly disjointed argument that people are routinely shooting 1080HD video on cell phones and keeping a terabyte worth of the raw on their hard drives.

"Well look." You have no clue what you're talking about when it comes to video, video editing, or video storage, and you seem to think it is normal for people to whip out their cell phones, shoot the equivalent of 9 feature-length HD movies, and then store them on their pro-grade edit suites.

I'm done with this thread. The stupidity is approaching sufficient density to form a signularity. Get the "last word" in if you want - it won't change the fact that you're spectacularly wrong.

Still Cool

[lymond01]

45 MP photo to zoom into:

Dubai

Re:Still Cool

[Greymist]

Quick correction, that's gigapixel, not megapixel. Some quick math says you'd need 375 images taken with the new sensor to make that image.

Re:Still Cool

[$RANDOMLUSER]

Bleh. Flash. Yuck. They should have used JPEG for that 45 GP photo.

Re:Still Cool

[Amanitin]

Budapest 70 gigapixels http://70gigapixel.cloudapp.net/index_en.html

Programmable robotic legs multiple lens

[h00manist]

I have a feeling that GPS and software integration to create auto-3d model photos are going more important than the resolution.

150 megapixel

[Anonymous Coward]

Good film under ideal conditions can handle 2500 line pairs per inch. The mathematical purist who was more obsessed with numbers than practical applications would want a sensor that can handle 10,000 dots per inch for copying film, and an image sensor of 5,000 dots per inch for shooting, with optics, electronics, and other hardware (and software!) to match.

5,000 dpi on a standard 35mm 3:2 aspect ratio means 37.5 megapixels.

For what it's worth, 10,000 dpi would be 4x that amount, or 150 megapixels.

Re:150 megapixel

[Anonymous Coward]

I've been a photographer for 40 years, have used both film and digital, and have lectured on photographic principles so I have to point out that this is not the whole story by far. Your film may only resolve 45-100 line pairs per mm, but that doesn't mean it's comparable to a 37.5 Mpixel digital array. There are around 500-3000 million silver halide grains per square cm of film, randomly distributed. Each of these is a photo site so there are on average around 4000 per linear mm. The digital sensor is a strictly rectilinear array of photo sites at the notional linear pixel count of, say, 208 per mm, and it takes a minimum of two photo sites to resolve a line. That means film has around 20 times as many photo sites per linear mm. So although both only resolve about 100 line pairs per mm, points in the image can be positioned with around 20 times more accurately on film.

On top of this, the digital camera's Bayer filter matrix inevitably reduces resolution further as it performs some averaging function (however mathematically sophisticated) over some three or more adjacent photo sites in each direction. So in reality, the digital array is more likely to require 3 to 4 pixels to resolve a line pair with reasonable contrast. That means a film-comparable digital array would need to be 400x400x24x36 or 138 megapixels. We get away with much lower resolutions in digital because most photos are fairly lacking in critical fine detail and modern colour repro techniques range from 72 (screen) to 600 (high end print) dots per inch (2.8 to 24 dots per mm). So your 37.5 Mpixel camera would yield a 10x8 at 600 dpi, but that doesn't mean it would have the same information content as a 10x8 from film.

Re:Suggestion...

[adonoman]

Yeah, no kidding - everyone knows that the Nikon D3X is WAAAYYY better than any crappy Canon.

Size doesn't matter

[$RANDOMLUSER]

I have to go with Ken Rockwell on this one: Megapixels don't matter. Unless you're blowing your 35mm shots up to poster size, pixel density over about 8 megapixels is useless overkill.

Re:Size doesn't matter

[Anonymous Coward]

If your name is Decker, and you want to see who is in the reflection of a curved mirror in photo, you're gonna need a lot of resolution.

Re:Size doesn't matter

[Bryansix]

That article is OLD and he is not saying that Megapixels don't matter. He is saying that to see a difference you need to quadruple the megapixels and also that other things matter a lot like light sensitivity, pixel to space ratio, ISO performance and the like. He then goes on to say you would need a 25 megapixel camera to meet 35mm uality and that such a camera is not feasable. Well I have to give him a Bill Gates because it is moronic to say anything is not technically feasable because in 10 years you look like a fool.

To get to the POINT, I own a Canon 5D Mark II which is a 21 Megapixel sensor. I have shot plenty of 35mm film and I can tell you without a shadow of a doubt that this sensor blows 35mm film out of the fucking water! You can see the images I take here. http://shezphoto.zenfolio.com/ and www.shezphoto.com Those are not even full res (although you can buy some of them full resolution). I have blown up the images to 24" x 36" and all the detail remains intact. I'm sure I could go larger but I just haven't.

Re:Size doesn't matter

[wiredlogic]

It all depends on how you intend to display the pictures. For most consumer applications the megapixel battle is over. If the pictures are only going to be seen on a screen or printed to something small like 4x5s, any modern camera will suffice. With an 8MP camera I can get acceptable prints up to 8x10 with just the slightest pixelation visible under close scrutiny. I recently had to shoot a picture for a book cover that I wanted to wrap from both extremes front to back across 15.5". This results in a tolerable ~200dpi using an 8MP camera whereas 300dpi would be preferred. I had to be careful with framing the shots because I didn't have the luxury of cropping or rotating them afterward.

Re:Size doesn't matter

[Anonymous Coward]

Anything Rockwell says is half true at best. He's a landscape photographer, so he doesn't do the kind of aggressive cropping you'll see in wildlife and bird photography - indeed, he tends to completely skip over the fact that other types of photographer even exist. Even for landscapes, certain types of print (such as cropping a wide shot to a panoramic format) demand more raw pixels than others. Around 6-8 megapixels is fine if you're printing uncropped images to be viewed at "standard" distances. For everything else, more pixels are nice to have.

Having said all that, this sensor isn't going to be in digital cameras any time soon. If it's released at all, you'll see it in astronomy or medical applications first. More likely, it's just a proof-of-concept thing.

Re:Size doesn't matter

[danpbrowning]

Ken Rockwell is to photography what a goatse troll is to Slashdot. (In fact, if you read his alien abduction pages, you'll see some similarity with goatse).

It's like saying "Computer specs don't matter. Unless you're folding proteins, a 486 is just as good as i5." While it's true that sharpness and resolution are not the most important factors in a photograph, it's misleading as their benefits do in fact contribute to most styles of photography, just as a faster computer can contribute to a better experience for most computing needs.

For example, most people feel that for an 8x10, there is no benefit to pixel counts above 6 MP, but in fact it takes a 24 MP before all the possible gains are realized, most importantly counteracting the loss in contrast from the anti-alias filter. (Many more MP would be required to hit full color resolution at Nyquist, but few natural images benefit from that, despite what the Foveon advocates claim.)

Re:Size doesn't matter

[kurokame]

It probably matters if you have an order of magnitude increase. Pixels in the sensor don't have to translate directly to pixels in the final image. If your target final resolution is significantly smaller than your initial resolution, this tends to translate to a significant increase in image quality.

Also, loads of people print poster-sized images (not that you need 120MP for a poster, and not that 35mm is directly relevant). Ever hear of this thing called advertising?

Re:Size doesn't matter

[thegarbz]

Unforunately this sensor also blows many of the lenses Canon makes out of the water. Wake me up when glass gets perfect and the laws of nature w.r.t difraction are broken. Until the ... 120mpx *YAWN*.

Re:Size doesn't matter

[darrylo]

Well, while megapixels don't matter beyond a certain point, the other features in newer cameras may make "higher megapixel" cameras better: better optics, better autofocus algorithms, lower noise sensors (maybe, but is often worse with more pixels), better in-camera raw-to-jpeg algorithms, etc., etc.. My newer, higher megapixel cameras certainly produce better-looking pics, out-of-the-box.

Re:Size doesn't matter

[djdanlib]

Or, perhaps, a lens and/or sensor that corrects for the distortion caused by the curved mirror?

Hey, they do it for the big telescopes.

Re:Size doesn't matter

[Bryansix]

I know photographers who only buy prime L series lenses. If you are willing to pay $1500 - $3000 for each lens and carry a bunch of them or walk to compose a shot then good glass can be had.

Re:Size doesn't matter

[thegarbz]

Yep but 21mpx even outdoes some (not all) of the the L series lenses. Just because it's got a red ring on the front of it doesn't mean some of the designs aren't lemons when subjected to this kind of scrutiny. Lots of L lenses way out performed the sensors of 3 years ago and no one was the wiser.

Yet what we're talking about here is a resolution that goes beyond the laws of nature. The airy discs generated simply by diffraction by a theoretically 100% perfectly sharp lens would be larger than the pixels on this sensor at about f/1.6 or smaller. That means regardless of how good any lens manufacturer can make a lens, it is physically impossible to match the resolution of the sensor for a zoom lens, and we all know that lenses are least sharp wide open so it's highly bloody unlikely that you'll ever manufacture a lens with this kind of resolution.

I am quite an optimist and very keen to see what the world will be like in 10 years, but while cameras have changed in simply astounding ways, the principles of the optics have remained the same since the begining with the only real technical advancements being in the improvement and speed of the electronic components (focus, and IS specifically).

Re:Size doesn't matter

[Bryansix]

All true. Everything gets complicated when you factor in the airy discs. Actually, even mentioning it in even a photography class will just get you mostly blank stares even from the professor.

Re:Size doesn't matter

[hazydave]

You're certainly up there... it depends on the film. The 5D sensor delivers a maximum hardware resolution of 78lp/mm. That's certainly up there, at least as far as sharpness goes (dynamic range, color accuracy, I'll leave to another discussion). It's a bit hard to compare to film, since you typically rate a film based on an MTF curve. If you don't have the curve handy, you can probably find f50 numbers, which is the resolution at 50% MTF. This puts T-Max 100 at 125lp/mm, Fujicolor Superia 100 at 63lp/mm, and Kodachrome 64 at a whopping 36lp/mm. Obviously, the 5D's sensor is 100% at 78lp/mm... so yeah, very sharp. And most chromes are similarly low in sharpness.. they're highly regarded due to fine grain. And no additional convolution in the printing process.

So yeah, in practical terms, the 5D is going to outperform any color film on sharpness, and pretty close to any pan film, if you can find one still made. This is not exactly a shock... there's a reason pros have largely moved to digital. It's not just consumers anymore.

here's a possibility I've often thought about

[Gnaythan1]

Hook this sensor up to a round lens and capture full 360 degree video all the time, and use software to un-distort the image so you have a fixed tiny camera, that you can pan and zoom all the way around with.

Re:here's a possibility I've often thought about

[ZenShadow]

Been there, done that, believe it was patented by iPIX. Not sure who holds it now since they're gone AFAIK...

Seriously, they used this to do those 3D virtual tours.

Uses

[MBGMorden]

I'm sure the professionals would love such a critter, but as a person who likes to just take personal stills, to me the megapixel war is over. At this point in time I have a hard time getting excited over anything more than 10-12MP. They print just fine to photo sizes that I'd be interested in, and the truth is that MOST of my photos I keep digitally anyways where anything that has more pixels than my monitor is a waste (particularly with the ballooning size of these photos).

I'm far more interested in seeing higher quality photos within our current megapixel options than seeing that particular number go up and up - afterall, there's a HUGE difference between your typical DSLR at 10MP and a $100 point and shoot at 10MP. That metric doesn't define the quality of the image.

Re:Uses

[that IT girl]

Yes--I am troubled that some industries (not just cameras, and consumers are just as guilty) are forgetting the old "quality over quantity" thing.

Re:Uses

[Anonymous Coward]

professionals don't want this, because they know how to take the picture they want to begin with, instead of relying on cropping out the middle 20% of a photo.

also, professionals know the noise tradeoffs, diffraction limit issues, mean the benefit of extra MP is a (very quickly) diminishing return.

Parent should be +1, Funny

[falzer]

You spell it Cannon and you're telling someone who shoots with a 1D and has likely used Zeiss lenses that they can't buy awesome Canon lenses in the toy department at Best Buy.

Re:Parent should be +1, Funny

[djdanlib]

I know, I was also amused. While writing about professional camera equipment, he spelled Nikon completely wrong :-D

"Oh come on, it's an easy typo. The keys are right next to each other"

Re:Suggestion...

[Anonymous Coward]

Are you joking? Think of the porn, man! I can see that girl's crabs so clearl--eek.

Resolution matters for serious cropping

[davidwr]

I'd love to be able to take many of my old family and vacation photos and take a small piece and blow it up to 4x6 or even 8x12 size without noticeable-to-the-casual-observer loss of detail.

Imagine taking crowd-scene photos of a sporting event then when your friend said he was there and points his face out in the crowd, you can print him out an 8x12 of him and his friends.

Re:Resolution matters for serious cropping

[$RANDOMLUSER]

Imagine taking crowd-scene photos of a sporting event then when your friend said he was there and points his face out in the crowd, you can print him out an 8x12 of him and his friends.

Even if your sensor could (theoretically) do that, your (hand holdable) lens couldn't.

Re:Resolution matters for serious cropping

[Entropius]

The loss of detail in those photos is because of optical defects, camera shake, and focus errors, not a limitation of the recording medium.

Re:Resolution matters for serious cropping

[Ichijo]

Even if your sensor could (theoretically) do that, your (hand holdable) lens couldn't.

With or without optical image stabilization?

Re:Resolution matters for serious cropping

[beh]

To me that sounds like you don't want to think about photo composition when you shoot - just give me a humongous multi-megapixel image, and I'll crop what I want.

It'll do the same as many zoom lenses do - it will stop many people about what makes a photo good - you just zoom the lens to make the subject cover the photo without even trying whether another vantage point may make the photo more interesting.

Try this some time - take a prime lens when you go out taking photos, it will force you into thinking about vantage points differently, because you may not easily get close enough to something, or not far enough away.

That said, most of the time, I DO have a zoom lens (70-200mm) on my camera - it is a more convenient for a quick snap here or there, but more often than not, I do put on a lens that doesn't lend itself to what I'm most likely going to see on a day trip, just to force me to think differently about what there is to see...

Re:Resolution matters for serious cropping

[lahvak]

It really has little to do with image stabilization, nor with hand vs. tripod. It has more to do with the behavior of light. Because of diffraction, you will never be able to under certain pixel size. You may be able to use deconvolution to get some additional detail, but with a small small sensor, you will probably never be able to crop that much. You would need to increase the sensor size. That said, having an affordable 100 megapixel large format camera would be sweet.

Re:Resolution matters for serious cropping

[Ichijo]

Because of diffraction, you will never be able to under certain pixel size.

According to the Rayleigh diffraction limit, diffraction is only an issue at small apertures. Using the lp/mm=1600/N rule, where N is the f-stop, a lens at f/8 (where optical aberration is usually not an issue) can resolve 200 lp/mm, which is 400 pixels per mm. On a 35mm sensor (36x24mm), this equates to 138MP. But a pinhole camera at f/200 achieves only 16 pixels per mm, which is 221,184 pixels on 35mm.

For the APS-H (27.9mm × 18.6mm), 13,280×9,184 pixel sensor mentioned in TFA, the smallest aperture that can be used before diffraction becomes a concern will be f/6.7.

Re:Resolution matters for serious cropping

[hazydave]

Right at the Rayleigh limit, MTF is about 9%. So using that rule, at f8, you're getting a theoretical maximum of 9% MTF at 200lp/mm. In other worse, you see something every so slightly different than a grey blur... in layman's terms, you have 9% lines, 91% greyness. You'll need to go to f2.8 to get a 50% MTF at 250lp/mm... that's going to look pretty decent with this sensor, though of course, you have half the original contrast. And these of course assume a theoretically perfect lens, which you don't have the in first place. Diffraction loss is only one source of trouble.

Re:Resolution matters for serious cropping

[hazydave]

I suppose it's historically correct to say that diffraction is only an issue at small aperatures, but that's largely a function of historical media -- film. Kodak publishes their resolution numbers, and the sharpest 35mm film they make is T-Max 100, which has f50 = 125lp/mm (50% MTF). Color films, and particularly chromes, are just not that sharp... ISO50 Veliva (which everyone shoots at ISO40) has an f50=45lp/mm. Kodak Ultra Color 100UC delivers 72lp/mm.

Given that the effective resolution will never be higher than that of the weakest link, using film, or most digital cameras, yields the observation that diffraction only matters at lower f-stops. But drop that 230lp/mm sensor in there,and things change.

Progression of film speed

[davidwr]

Anyone remember when ISO 32 was "fast"?

I miss Kodak Royal Gold 25. Yes, there are analog films that are "just as good" but they are not "just the same."

Re:Progression of film speed

[djdanlib]

Ah, the good old days... I wasn't around for it, but I wish I was. I can see the Kodak plant that used to manufacture it from my office.

What are your favorite films that are still manufactured?

Higher quality photos...

[davidwr]

Require more discriminating photographers who take the time to learn what makes what most people would call a "good" photo, plus equipment that makes it easy for them to take those photos without blowing their budget.

Automatic modes that "do the right thing" with most scenes go a long way, but that's still no substitute for good composition and knowing that if you wait 5 minutes for better lighting or for the car that's blocking your subject to move out of the way you'll get a better photo.

Post-processing is your friend

[davidwr]

I'd rather have a "raw" image of 120 megapixels with 16 bits per pixel that I could post-process than a 30 megapixel with 64 bits per pixel.

I can post-process the former into the latter but not vice-versa.

Problem with this....

[Obliterous]

at that resolution the pixel sensors are closer together than the wavelengths of visible light, and each photon will be triggering multiple pixels, thus reducing the apparent resolution.

Re:Problem with this....

[Anonymous Coward]

Let's see. Red-Orange light from a HeNe laser is 632nm. 120,000,000^.5 ~= 10954x10954, so if the sensor is less than 6.9mm square, it might have a problem with that color. Since the sensor is probably the size of 35mm film or larger, this isn't a problem

However, I don't think one photon *can* trigger multiple pixels, since light is quantized, the best you can do is 1 to 1, with most pixel sensors requiring several photons to trigger.

Re:Problem with this....

[fbjon]

It doesn't matter what one photon does, since on average, many photons will "fill up" the probability space. Photons originating from one point will hit one sensor site with some probability, and the sites next to it with some lower probability, and that distribution ends up being the blur you see. Supposing the sites are that tiny, of course.

You're right about the sensor not having this problem, though it's slightly closer than you indicate:
An image on 135 film is 36x24mm, and APS-H has a crop factor of 1.3, so we get: (24mm/1.3)/9184 pixels = 2010 nm per pixel, which is 2.6 times the lowest wavelength (760nm or so).

Definitely need better physics

[delta407]

A more substantial problem is that diffraction limits the effective resolution of an optical system to well above the size of each of these pixels. This is a problem with current sensors at narrow apertures; lenses exhibit a measurable loss of sharpness, typically f/11 and up, because the airy disks expand as the aperture contracts. With hugely dense sensors like this, though... plugging some numbers into a website that explains the whole situation suggests that you'd need to shoot with apertures than f/1.8 to get circles of confusion smaller than the size of a single pixel.

That's right--even "fast" f/2.8 lenses are limited by physics to never being able to project detail onto individual pixels. You could potentially add a deconvolution stage in software to recover additional sharpness, but not in hardware.

Another thing. Do the math: the pixels are 2.1 micrometers square. Compare to trichromatic human vision, which detects red light peaking at 564 nanometers, 0.564 micrometers. The size of a pixel is within a factor of four of the wavelengths they measure. Staggering.

Glass isn't the problem. We need new laws of nature, since we're near the edges of the ones we have now.

Re:Definitely need better physics

[danpbrowning]

That's right--even "fast" f/2.8 lenses are limited by physics to never being able to project detail onto individual pixels.

That is incorrect. Parity between the airy disk and pixel diameter is not the point at which additional detail becomes impossible -- that is only one point on the curve of diminishing returns. In other words, it is the difference between the "diffraction limited" spatial frequency and the "diffraction cutoff" spatial frequency. It is only the latter that denotes the impossibility of further resolution from decreased pixel size.

The easiest way to understand this is to look at MTF. When diffraction causes the optical system MTF to drop to 50%, most would consider that the end of the line. But in fact, that is just the point where a lot of contrast is lost -- detail is still there and contrast can be restored with sharpening (e.g. RL deconvolution). MTF must drop to 10% before detail truly becomes extinct, and for a 2.2 micron pixel like this 120 MP Canon, f/5.6 will still give you 18% MTF, and there are a host of lenses that are very sharp at f/5.6.

For further consideration, look at the effect of the anti-alias filter, which drops MTF of spatial frequences far lower than needed to suppress aliasing. The ideal solution to this problem is pixels that are so small that diffraction itself anti-aliases. That will increase contrast at lower spatial frequencies by 30%.

Re:Definitely need better physics

[dangitman]

We need new laws of nature, since we're near the edges of the ones we have now.

That's OK. Once we figure out the meaning of Life, the Universe, and Everything; a new universe will be created to replace the one we had.

Not today....

[davidwr]

I have faith in the future of technology....

I wouldn't at all be surprised if image stabilization becomes cheap enough to put in hand-helds.

Eventually, image sensors will be small enough that the lenses will contain very little optics and it will be easy to make them cheap and sharp.

With "liquid lenses" that are electronically reshaped on-demand and in real-time, we might see the day where every shot is technically dead-on sharp, almost to the limits of the laws of physics. Such a day may not be in my lifetime but it will happen if there is a market demand for it.

Re:Not today....

[Anonymous Coward]

Image stabilization already is in hand-helds.

Re:Not today....

[fbjon]

The trouble is with the limits of physics, rather than with the engineering...

Re:Not today....

[Carnildo]

With "liquid lenses" that are electronically reshaped on-demand and in real-time, we might see the day where every shot is technically dead-on sharp, almost to the limits of the laws of physics.

We hit the limits of the laws of physics decades ago. High-quality lenses have been diffraction limited since the mid-60s for non-zoom lenses, and since the late 90s for zoom lenses.

As an example, at an aperture of f/8, no lens, no matter how good, can project a point of light to cover less than nine pixels on the sensor described in the article. At f/22 (the standard for high depth-of-field photography), a point of light will cover approximately 65 pixels. At f/32 (needed to get a decent depth of field on a long telephoto lens), it'll cover about 133 pixels.

Re:Not today....

[Dogtanian]

I wouldn't at all be surprised if image stabilization becomes cheap enough to put in hand-helds.

Er, it's already in handhelds- even relatively inexpensive ones like this circa-2009 Canon I bought as a present a while back.

Re:Not today....

[hazydave]

Seriously. I bought my wife a Panasonic Lumix P&S digital 3-4 years ago, with optical image stabilization. She had been using my Canon Pro90IS, also with optical image stabilization... that's from about 10 years ago. Most every consumer P&S and camcorder has some form of image stabilization these days.

But this has nothing to do with diffraction. You can't beat physics. Small cameras have been very affected by diffraction for a long time (including video cameras)... it's less of an issue in the 35mm-class world, but still an issue. When you stop way down, your sharpness is going to be diffraction limited. Even back in the days of film.

If all

[Mathness]

If all you care about is resolution (in a SLR camera) this is great. But if you also care about stuff like low noise, dynamic range or a diffraction that isn't limited to f/4, you are much better off with at least a medium format camera.

Sci Fi Cliche

[mauriceh]

Hmm, with that resolution we could do the science fiction standard nonsense:

"Select quadrant in top right corner. Enhance.
Select the reflection on the subjects glasses. Zoom 50X and enhance.
See the face of the murderer.."

Remember Blade Runner?
http://criticalcommons.org/Members/ironman28/clips/bladeRunner3DphotoH264.mov/view

Re:Sci Fi Cliche

[Anonymous Coward]

hehe no I think RD did it best

http://www.youtube.com/watch?v=KUFkb0d1kbU

Re:Sci Fi Cliche

[PPH]

Why go back that far? Every CSI show has a bit where the IT person takes a security cam photo (in real life, a bank robber's face occupies 6 pixels) and blows it up to read a license plate number across the street.

Re:Sci Fi Cliche

[mauriceh]

The Blade Runner was both the oldest example I could remember, AND it was the most far fetched version.

Re:Sci Fi Cliche

[grumbel]

Hmm, with that resolution we could do the science fiction standard nonsense:

The fascinating part of that scene is that it actually is extremely close to reality. We already have tons of gigapixel images floating around on the net and in terms of resolution they seem to be quite up on par with the Bladerunner image (i.e. 10 gigapixel or so). The Bladerunner image gets a bit further in that it is not only 2D, but actually a bit 3D, but even that is possible with lightfield photography. Now today those gigapixel images are produced by cameras mounted on robots, but when you look at Sweep Panorama it is not hard to imagine that in a few years down the road we will be shooting those images with regular consumer gear. The final issue one might complain about is that he scans what looks like an actual analog photo for all that, but it is not hard to imagine that the picture itself might not contain all that data, but instead contain it on an embedded chip (see MicroSD) or the printed picture could just act as key to the full-res picture stored on the cloud. Or of course, maybe print resolution just got better, after all a 10 gigabyte image should fit nicely on a BluRay and thats about the same size as a photo.

Re:Sci Fi Cliche

[grumbel]

And an additional throw in: Does anybody know of gigapixel images that capture mundane stuff? Cars, people, etc. instead of those large scale panoramas (recreation of the Bladerunner picture would be perfect of course)? The closed gigapixel images I could find that are not city panoramas are www.gigamacro.com/, i.e. extreme closeups of money, insects and other stuff.

The real application is surveillance cams

[Animats]

The real application for ultra-high resolution is surveillance cams. Something interesting might happen somewhere in a wide field of view, and when it does, detail is useful.

Re:The real application is surveillance cams

[joh]

The real application for ultra-high resolution is surveillance cams. Something interesting might happen somewhere in a wide field of view, and when it does, detail is useful.

You may have a point here. With something like 120 MP and a wide lens you can cover a large area with one camera and have access to details everywhere without having to move the thing around. On the other hand you get absolutely *huge* amounts of data...

Re:The real application is surveillance cams

[dangitman]

The real application for ultra-high resolution is surveillance cams.

Because, as we all know, there can only ever be one real application for a particular technology. The multitude of other valid applications are just faking it.

Re:The real application is surveillance cams

[ZERO1ZERO]

Obviously CSI have been using them cameras for years..

ENHANCE!

yeaaaaaah!

Re:The real application is surveillance cams

[Animats]

On the other hand you get absolutely *huge* amounts of data.

There's an opportunity there for developing new compression techniques. Compression for surveillance data really should be quite different from entertainment content. You want a really high quality frame once in a while, maybe once a second, from which stills can be extracted. You can accept considerably lower quality for intermediate frames. Compression of scenes that don't change much should be very high. An option to go to higher quality when something interesting appears or some other alarm system indicates activity is worthwhile.

It matters for me...

[PhantomHarlock]

From a professional photographer's standpoint, I DO appreciate more resolution, because I do make things that end up on posters and billboards. Also, the primary advantage in most cases is the ability to crop and still have a decent resolution image.

As another poster mentioned, the main problem at this point is with the glass. Sharp glass that remains the same size to accommodate a denser, not larger sensor is a tough proposition, and the new frontier of technology. Things like liquid lenses may overcome this in the future, who knows.

Right now, with my 21MP 5D Mk. II, I can use modern Canon "L" zoom lenses too my heart's content and have an image that is sharp from corner to corner, especially now that you can easily correct for chromatic aberration in RAW processing software. (to give you an idea of how far this has come, when I was doing 3D animation 10 years ago, we would commonly add back in chromatic aberration to 3D generated images to give them a sense of realism.)

For the sort of resolution discussed here, if you wanted relatively sharp pixels at 1:1 (spatial, or perceived resolution, actual sharpness delineation from one pixel to the next) you would probably want to stick with prime (non-zoom) lenses with fewer glass elements, and it would probably OK.

Other posters are correct in that this kind of resolution is currently unnecessary for consumer and casual use. But for me, large blow ups and two-page spreads are a frequent thing, and I apprecicate all the pixels I've got. :)

Remember Blade Runner?

[Anonymous Coward]

Yes.

But I'm beginning to wonder if it is just a Tyrell implant.

535mb images?

[pointbeing]

Did a bit of math here and at 36-bit color a raw image would be a bit more than 535mb.

I don't think the technology is available yet to process an image that large into a jpeg or copy a raw image to a storage device quickly enough to use this in most camera applications - and definitely not in your point and shoot ;-)

Re:535mb images?

[LordKronos]

Currently used sensors don't capture RGB for every pixel***. Each pixel is one color, and then processing is done to interpolate the other colors from adjacent pixels of those colors. Go look up "bayer filter" for more details.

So really, rather than storing 36 bits per pixel, you'd only be dealing with 12 bits. Actually, most likely more than that. Current Canon SLRs capture 14 bit per pixel. But the point is you don't need 36 bits. Lets just say 16 bits, which gives you 240 MB for a 120 MP sensor.

On top of that, there is also compression. When stored in raw file format, there is lossless compression on the data, so it would come in somewhat less than that. My 12 MP Canon XSi is 14 bit per pixel, so you'd expect the raw files to be over 21 MB, but they typically end up being 12-18 MB. Then there's JPEG, which would only store 8 bits per pixel, and compress much more (though lossy), so you'd expect well under 100MB (probably more like 50 MB)

*** There are some sensors that do this, but they were only used in a few less popular camera models, all of which (to my knowledge) are discontinued.

Re:535mb images?

[danpbrowning]

This is a solved problem. RED ONE has proven that software can scale file size linearly with quality so that there is no file-size or computing speed disadvantage to larger pixel counts. (Not to mention that the typical balance between quality and size is almost an order of magnitude worse than what it could be.) Now if only the rest of the industry would start paying attention to the software engineers...

Re:535mb images?

[pointbeing]

I stand corrected. Thanks for keeping me straight ;-)

Re:535mb images?

[fyngyrz]

There are some sensors that do this, but they were only used in a few less popular camera models, all of which (to my knowledge) are discontinued.

Nope.

The future is now

[freelunch]

boasts a ridiculous resolution of 13,280 x 9,184 pixels

My 6x7 cm film images are already 11,023 x 9,448 when scanned at 4000 dpi.
And there are no artifacts from Bayer interpolation.

30x36" prints, and even larger, are spectacular. But you need good lenses, a good tripod, and good technique; otherwise you won't resolve the detail.

And with 20x30" prints only $9 at Costco (on profiled printers), I *am* enlarging my prints to poster size, thankyouverymuch.

I look forward to digital catching up.

Re:The future is now

[falzer]

> I look forward to digital catching up.

It's slowly getting there. PhaseOne has a 60MP medium format digital back, for example.

BetterLight has a digital scanning back for large format (no bayer interpolation artifacts either) which gives ~139MP images. Admittedly it is not like the sensors we know and love in regular digital cameras.

Re:The future is now

[Anonymous Coward]

It's slowly getting there. PhaseOne has a 60MP medium format digital back, for example.

Yes, though at $40K-$60K, they hardly deserve mention.. Plus there are the concerns about service and longevity. Progress is being made on the high end, but verrry slowly. Pentax 67 bodies can be had for a couple hundred. My Nikon 9000 scanner was $1900, and is apparently still worth what I paid, or even more. So, at about a buck a shot for film and processing, and a dslr to supplement for "spray and pray", I'm happy.

And yet, somehow, an 8x10 piece of film contact printed looks absolutely amazing; at just 8x10. We have a long way to go with digital.

Re:The future is now

[dangitman]

My 6x7 cm film images are already 11,023 x 9,448 when scanned at 4000 dpi.

Just because your scanner is capable of resolving 4000 dpi doesn't mean that your film is resolving detail at that resolution. Most likely you are wasting pixels, and think you have a higher resolution image than you actually do.

Re:The future is now

[Anonymous Coward]

Thanks for the speculation. The neat thing about film is that you can look at the original to see what is actually present, and compare it with what you are resolving in your scans. So, really, there is no mystery. And then, of course, there are the 40x60" enlargements, and smaller prints.

Also, in comparisons between 4000dpi and 8000dpi, 8000dpi is an advantage. If I could afford an 8000dpi scanner, I'd have one.

Re:The future is now

[dangitman]

The neat thing about film is that you can look at the original to see what is actually present, and compare it with what you are resolving in your scans. So, really, there is no mystery. And then, of course, there are the 40x60" enlargements, and smaller prints.

None of which prove anything unless you subject them to technical/scientific tests. Just because you think they look good doesn't prove anything about the actual resolution.

Re:The future is now

[Anonymous Coward]

None of which prove anything unless you subject them to technical/scientific tests.

A microscope measures lp/mm from the actual film. What more is there to "prove"?

Comparing a 4000 dpi scan against an 8000 dpi "proves" that the 800 dpi resolves more detail.

There are plenty of folks on the internet who spend their time nerding out on tests. You might want to check out those forums if you're into that sort of thing. For me, the goal isn't numbers, but rather high quality prints at very large sizes. The digital camera resolution just isn't there yet.

Light Field Camera

[cowtamer]

I'm sure it'll be perfect for this application:

http://en.wikipedia.org/wiki/Plenoptic_camera (a type of camera that can let you re-focus (and to a certain extent re-position) images after taking the shot. The problem is that it requires a LOT of resolution to produce acceptable images).

http://graphics.stanford.edu/papers/lfcamera/

http://www.youtube.com/watch?v=9H7yx31yslM&NR=1 (demo video from paper above)

http://www.youtube.com/watch?v=o3cyntPC2NU

Here's one built with a 250 MP Flatbet scanner:

http://www.youtube.com/watch?v=4O5fPoacF3Q&feature=related

50 megapixel cameras

[Anonymous Coward]

Canon unveiled a 50-megapixel sensor in 2007, but that's not made it any further than the labs to date."

True, but you CAN buy cameras with that sort of resolution from other manufacturers, such as Hasselblad. In fact, Hasselblad has 60 megapixel models available, too.

Not cheap telephoto

[jridley]

About 21 megapixels on a full frame SLR is already pushing the resolution limit of reasonably priced lenses (IE, L series glass). You might get a bit more than that, say 30 megapixels. Beyond that you're exceeding the Dawe's limit of the optics, and you're just not going to get any more detail this way than by just interpolating the digital 30 megapixel image.

Re:Not cheap telephoto

[Arimus]

*Some L Series Lens, some are somewhat expensive for semi-pro photographers... like this beasty... 800mm f/5.6 a snip at £15k.... http://www.canon.co.uk/For_Home/Product_Finder/Cameras/EF_Lenses/Image_Stabilization_Lenses/EF_800mm_f-5.6_L_IS_USM/

Sorry, still somewhat lame

[Tri0de]

I used to use an 8"x10" camera, with 25 ASA film.
As much as I really like digital, and I do, there is simply no way an 8x10" ('contact')print from a mere 120 megapixel file is going to be even close.
I'll get stoked when we're talking 100+ gigapixels.

Re:Sorry, still somewhat lame

[Carnildo]

Why are you comparing large-format film to 35mm digital? You should be comparing like to like, and looking at large-format scanning backs.

Re:Sorry, still somewhat lame

[Tri0de]

Because the article was about the Canon 120 megapixel chip.

Re:Sorry, still somewhat lame

[Failed Physicist]

Yes, a 120 megapixel, 35mm (24x36) chip. That is, by the way, about 14 000 lines per inch. 59.7 of those sensors would be needed to fill a 8"x10" frame. So at this pixel density, what you thought was a "mere 120 megapixel" actually becomes a 7.168 gigapixel image.

That is an absurdly high number; printing at 300 dpi, (assuming square image for lazyness), that would make a print 24 feet across.

But who looks at a billboard with a magnifier?

Only use I can conceive (though there are, of course, probably more) is using a surveillance camera with, say, a 106 degree FOV (16 mm lens) and crop 440x into it for a 500 mm FOV (5 degrees) with an end result close to a modern camera - 16 mp. But this would require lenses far above and beyond current top of the line, professional lenses: current pixel densities in DX cameras (that are much more dense than 35mm) are already able to resolve more than even the center (which is the part with the best resolution) of modern lenses can show. Measurements go around 2000-2500 lines per inch for golden-ring Nikon and Canon L series, a number which, by the way, fits perfectly well with current pixel densities (Canon is already pushing it a little at 24 mp, and next-gen 40mp will already need a refresh of the L series to perform optimally). Referring back to the 14k lines per inch for the sensor above, you can see how much would be wasted without any currently existing lens that can resolve that much.

For the record, I shoot with a 12mp D700. Not as good low-light as a D3s (which is about 2 stops better) yet I am still amazed that I can shoot handheld after sundown without any noise or even by moonlight in the dead of night if I'm ready to compromise a little bit on noise (6400-12800 iso). I can't understand people who want more pixels than that; reducing noise in low-light situations is my main priority and more pixels are directly counter to that.

Re:Sorry, still somewhat lame

[dh003i]

There is nothing wrong with comparing different formats. There are things that you can do with a large-format 4x5 that you still can't even get close to doing with a full-frame DSLR, nor even really with a medium-format digital camera. But of course there are advantages for digital too.

There is really no direct comparison between 4x5 film and a scanner-back. The scanner back requires a much longer exposure, it is not a sensor. For something very static, it would work fine.

A 4x5in piece of film, assuming that 3.75x4.75in of it is usable, has 17.8125 sqin of area. A 36x24mm (1.42x0.94in) digital sensor has 1.3392 sqin of area. That's 13.3x the area and 3.55x the diagonal.

Many lenses for 4x5 will resolve close to the diffraction-limit (1600/32 = 50 lp/mm) at f/32*, maybe they'll resolve ~46 lp/mm, or 92% of the theoretical diffraction limit). f/32 on 4x5 is equivalent to f/9 on 35mm film (in terms of DoF at an equivalent angle of view & distance from subject, i.e., same framing). To match the resolution of 4x5, you'd need to have to come achieve 92% of the diffraction limit of 178 lp/mm at f/9: 164 lp/mm.

My Olympus ZD 50/2 macro lens, one of the sharpest lenses ever made, resolves "only" 123 lp/mm at f/4 (31% of the maximum theoretical diffraction limit). At f/9, it resolves 96 lp/mm (54% of the theoretical diffraction limit). Good luck finding a lens that resolves 164 lp/mm at f/9 and that is suitable for general-purpose use.

*http://www.hevanet.com/cperez/testing.html/

Dynamic Range,

[140Mandak262Jamuna]

I wish they would spend more time on improving the dynamic range than to just play the megapixel count wars.

Instead of total pixel count, get one set of pixels to shoot at the equivalent of 100 speed, and the adjacent set of pixels to shoot at 200 speed etc etc. Then process the pixels to get details in dark regions and to scale the brightness. I would like a dynamic range (brightness ratio of the brightest to dimmest pixel) to be a million or more, not the present 1000. Human eye has a dynamic range of about 1 million (only in the fovea, not in the peripheral vision).

Re:Dynamic Range,

[Anonymous Coward]

You seem to miss the point that an on-sensor HDR approach you suggest *requires* a higher total pixel count, because with that approach 3 (or more) sensor pixels are required to generate a single image pixel.

Re:Dynamic Range,

[danpbrowning]

I wish they would spend more time on improving the dynamic range than to just play the megapixel count wars.

Agreed. The pattern noise in their $2,500 5D2 limits the dynamic range to even lower than their cheap little G11. It's embarrassing.

Re:Dynamic Range,

[RichardKaufmann]

Actually, you'd could use this technology to give you amazing dynamic range. Let's say all you cared about was ~7.5 megapixels. Well, take the 120mp sensor and carve it up into 4x4 tiles. Assuming a Bayer array, 8 of these sensors will be green sensitive, 4 red and 4 blue. OK, now change the filters on these sensors so that they scale for different levels of light (not just wavelength). You now have four times the dynamic range in red and blue, and perhaps even more in green (you can trade off perceived resolution for dynamic range in the green sensors).

Re:Dynamic Range,

[bobaferret]

I think HDR is a strong possibility the way you describe it. There may be faster frame rates by sweeping one set of sensors then the next. They could add a whole new variety of color filters. Right now I think they sit at 4. I really think it has something to do with speed though given that it's on their 1D sized sensor, which is all about speed. I don't think it's about large format cameras, canon seems reluctant to really move into that area. I'm curious what the paparazzi and sports photographers are missing. Because that's their bread and butter. Canon doesn't like it when the sidelines of the big game are filed with anything other than big white lenses with little red rings and those are all APS-C sensors behind them. There's always 3D too, but I have no idea how this new sensor would help that.

120 Megapixels might be ok for a point and shoot

[edremy]

but serious camera users need at least 3.2 gigpixels to fully exploit a decent lens.

I admit, portability suffers a bit at this point, but aren't your pictures worth it?

Enhance!

[Anonymous Coward]

Zoom in there. Enhance. Sharpen it. Enhance.

Great combo!

[Ossifer]

I hope they package this behind a nice 3mm plastic fixed focus lens!

Re:120 Megapixels might be ok for a point and shoo

[Shag]

The only problem with that setup is that it takes pictures of what it wants to. ;)

I'll have to stick with the next best thing, which I at least get to point at things. :)

In other news...

[MasterOfGoingFaster]

In other news, Ford has set a new land speed record by attaching a Mustang to a solid-fuel rocket from the space shuttle. Funeral services for the driver/pilot will be held next week.

A sensor beyond 20 MP is of limited use - it out-resolves nearly all commercially available lenses. This is when professionals move up to medium format cameras and lenses to achieve a larger image area. Diffraction and noise are just of the few problems that have not been resolved with small dense sensors.

pixel size

[Anonymous Coward]

The pixel size actually isn't that small. The numbers in the article work out to a 2.2um pixel pitch which is fairly common for low end webcams. Of course, low end webcams have smaller overall area and much lower image quality.

Think the other way

[RickyG]

If you have the technology to make a 120 megapixel camera, reverse your thinking. Can you use that technology to decrease the size of your current product, so that a standard 8 to 10 megapixel camera is so small and compact, that it meets the needs of the growing phone/ipod/iphone/ipad industry?

Re:Think the other way

[jjohnson]

The size constraint for cameras is the lens, not the sensor size. We already have small, handy point-and-shoots that match the resolution of DSLRs. But the image quality is generally poorer because the lenses are smaller (and not manufactured as exactingly). A camera on an iPod Nano will suck because the lens will be tiny and collect much less light than a 35mm lens.

One interesting development is origami lenses that use a flat array of angled mirrors to collect the same light as a comparable-diameter lens, but without the length, so you can have 35mm lenses of any focal length that are an eighth of an inch thick--your iPhone can then be as good as a DSLR for image quality.

Re:Think the other way

[Anonymous Coward]

Yes, but the lens will still suck golf balls through a straw, and so the pictures won't actually be any better.

Re:Think the other way

[Khashishi]

Wow, this is the kind of stuff that deserves patent protection.

Physics is the ultimate limit to resolution

[Anonymous Coward]

Light diffraction limits even a perfect lens from having an infinite resolution: http://en.wikipedia.org/wiki/Airy_disk

I'm sure this part is for an industrial camera using specific wavelengths of light and wouldn't be practical for an everyday camera.

Not as useful as you'd think for the same reasons

[bytestorm]

What you suggest has the same problems this high pixel density sensor has. The physics of optics require a large area to get high performance out of those megapixels or else diffraction comes in to play and a given input pixel will get mixed up with all of the pixels around it on the sensor. On a small device like a phone or ipod, you don't get a lot of sensor area due to the size limitations of the device so this problem comes into play much earlier. This is part of the reason why you don't see more than a few megapixels on embedded devices.

quantity

[Anonymous Coward]

I believe it was Bill Gates who recently said 12MP should be enough for anyone

some daguerretypes have gigapixel resolution

[peter303]

They've been microscopically digitized and zoomed to wall size without loss of resolution. In one photo they discovered a clock in a far-off building that gave the time of photograph.

Re:some daguerretypes have gigapixel resolution

[Khashishi]

That's very interesting. Are there any images on the web? I'm surprised that old lenses were good enough for that.

Re:some daguerretypes have gigapixel resolution

[Failed Physicist]

pinhole cameras (from maybe f54 and going even higher than f1000) have essentially infinite depth of field, leading to perfect focus anywhere.
Diffraction got really bad over f200 though, but larger "film sizes" tend to counter this. Daguerrotypes are essentially the glass pane on which the image is projected, so they had a lot of magnification potential.

Re:Not as useful as you'd think for the same reaso

[RickyG]

I would hope that Canon, a big name in lens for years, would be working to overcome that limitation. It is hard to imagine that they wouldn't be doing something in the lens field to match and make the sensor usable. But, maybe as the report had stated about the 50 Megapixel not showing up, as you said, it is that difficulty they haven't overcome. It is like being given a $1000 bill and no one having the change to break it...

what I thought I was reading...

[Anonymous Coward]

"The sensor is so densely packed that during a standard exposure, only one-sixtieth of its receptors will be hit by a single photon."

A whiter shade of pale?

[spun]

So, you are saying that we can have a blacker black than black, and a whiter white than white? Because as I see it, averaging pixels in hardware gives you the ability to react to far fewer photons than it would take to activate one small pixel. Giving you greater range between the darkest pixel distinguishable from black to the lightest pixel distinguishable from white.

Re:A whiter shade of pale?

[spitzak]

Yes as I said below, averaging a lot of pixels would lower the noise floor and increase the range. However it increases the range by far, far less than if you used those N pixels for N different-exposed shots and this sort of huge range increase is normally what is meant by "HDR".

Re:A whiter shade of pale?

[spun]

Oh, well, yeah. Of course.

Why hold your breath?

[Khashishi]

The article poster is being very silly. The only use for this is for scientific instrumentation. It's useless on a consumer camera, even a professional one, because the optics aren't perfect enough for that kind of resolution to matter. Perhaps it could find use in a large reflecting telescope in an observatory.

The Megapixel Myth

[EmagGeek]

I'll take a 6MP sensor over a 120MP sensor, in the same sensor size, any day of the week.

120MP won't be that useful if you have to be aimed at the Sun to pick up any light.

Re:Not as useful as you'd think for the same reaso

[Khashishi]

Diffraction isn't the main problem. The main problem is insufficient photons. Smaller CCD pixels means they get less photons per pixel, and the signal to noise is lower. Cellphone cameras already get crap performance in low-light conditions, because their lens is so small and doesn't pick up a lot of light. Increasing pixels just gives you a more gritty picture.

For better cellphone camera performance, they need to work on increasing quantum efficiency or lowering noise (perhaps by actively cooling the CCD).

Telephoto and format

[muridae]

This sensor is APS-H, just a wide version of an APS-C sensor. 30mm by 16mm, roughly, just barely comes close to 135 film at 36x24mm and you will need six of the APS-H sensors to get close to 6x4.5 medium format. More sensors stacked is you want a larger medium format exposure, and a rather prohibitive number of them if you want something in the large format range. Frankly, with the noise level of that many sensors crammed into that little space, the benefit of getting a potential 720 megapixels out of a 6x4.5 camera is lost. A larger single sensor, in the 50 to 100 mp range, would have a much lower noise and get you a great photograph. Large format will be a whole other game. I would love to see a 4x5 sensor as a slide in replacement, but at that size you usually see scanner backs that can result in nice resolutions upward of 120 mp. You just have to have the cash for them.

As for telephoto; if cropping equals zoom, go for it. But, if you end up using just 12mp out of that 120mp sensor, wouldn't you have been better off using a 12mp camera, and buying a zoom lens? I have to guess that would work out to be cheaper.

You'd still get more detail, though

[Estanislao Mart�nez]

Remember that most sensors are Bayer array sensors, where each photosite records only one of the primary RGB colors, and the value for the other two channels must be interpolated from adjacent photosites. This means that even if an in-focus point on the subject is projected to a disc that's larger than an individual photosite, the spillover into adjacent pixels actually contributes to resolution, because it gives you more information to use at the demosaicing stage.

Put 4 photosites (RGBG) inside each circle of confusion and you may have built yourself something comparable to a Foveon sensor.

you'd need a lens capable of 232 lp/mm

[dh003i]

The APS-H sensor is 28.7 x 19 mm. To resolve 13,280 pixels along the length of the sensor, you'd need a lens that could resolve 463 lines per mm (232 lp/mm). According to the laws of diffraction, this is impossible for f-stops greater than approximately f/3.24. (1500/463 = 3.24). That doesn't give you a lot of depth of field to work with, if you want to resolve all of those pixels. And you don't have a bellows design capable of tilts and shifts, as do 4x5 large-format cameras, so that compounds the problem.

The practical problem right now is that there just aren't any lenses that resolve 232 lp/mm for normal photographic use. There are some very specialized lenses that deliver many hundreds of lp/mm. e.g., the 28 mm f/1.8 Ultra-Micro-Nikkor resolves about 600 line-pairs per mm or 1200 lines/mm: http://www.naturfotograf.com/lens_spec.html/. The conditions under which the lens resolves that many lp/mm are very limited, however (macro only, at a very specific magnification).

Re:you'd need a lens capable of 232 lp/mm

[dh003i]

PS: That 600 lp/mm 28 mm f/1.8 Ultra-Micro-Nikkor lens I mentioned has an image circle of just 8mm and is optimized for 1:10 magnification (e.g., it resolves an 80mm subject size on an 8mm sensor or piece of film). Reversed, it would be optimized for 10:1, but would only produce 60 lp/mm (600 lp/mm are still "resolved" on the 8mm object side (4800 lp total), but this is all that can be produced on the image side (4800 lp / 80mm = 60lp/mm).

So this is a very limited application lens to get that kind of resolution.

To expand on this...

[Estanislao Mart�nez]

I think this merits explanation in a bit more length.

Nearly all digital cameras have Bayer array sensors, where each photosite only records the value for one of the three RGB color channels. A 12MP Bayer array camera produces full-color images with 12 million RGB pixels, but that overstates the amount of information that the sensor captures by 3x; for each pixel in the resulting image, only one of the three channel's value was actually directly recorded from the scene, and the other two channels' values were interpolated from the values of adjacent pixels that recorded the missing channels.

Or, the quick way to put it, a 12 megapixels Bayer array camera is really 6 green megapixels, 3 red and 3 blue. This has several consequences:

1. The sensor is susceptible to color moiré artifacts at its resolution limit. To avoid those artifacts, typically there is an optical anti-aliasing filter in front of the sensor that blurs the image a little bit, so that some of the light that would have fallen on only one photosite is spread to hit adjacent ones. This comes at a resolution cost.
2. The effective resolution that you can get varies with the color of the subject. There's a good discussion of this effect at this page. But basically, if you're photographing a strong red or blue subject, your 12MP camera is closer to a 3MP camera.

These two things mean that you can get resolution improvements from putting more photosites on a Bayer sensor, even if the size of the individual pixels is smaller than the circle of confusion of the lens.

Imagine if the length of the side of the photosite coincided exactly with the diameter of circle of confusion. This means that a point on the subject that aligns perfectly with the center of a photosite is going to project entirely inside that photosite. Now assume that point of light is pure red. If the photosite is a red-sensitive one, the sensor then records the fact that the point has a strong red component, but it can't tell if it has a green or blue component. If the photosite is green-sensitive, then the sensor records the fact that the point has no green component, but it can't tell whether it has a red or blue component.

Now, however, imagine that the photosite is smaller than the circle of confusion. Then some of the light is spilling over to adjacent photosites--which means that you record a value for all three color channels for that point on the subject. This makes it easier to infer the values of the missing channels at the pixel that corresponds to that photosite, because the adjacent photosites will have recorded it.

So, making the pixels smaller beyond the lens' diffraction limit lets you (a) use a weaker anti-alias filter on the sensor (or none at all); (b) gives you more consistent resolution for subjects of different colors. If you go all the way, you'd make your sensor have 4x the amount of photosites as the number of pixels in the output images: e.g., you'd build a camera with a 60MP Bayer-array sensor but output 15MP images, using 4 photosites per output pixel (and no antialias filter). That would outperform today's 15MP cameras.

Wrong.

[Estanislao Mart�nez]

If it's a Bayer sensor, you do get more detail from making the sensor resolution finer than the lens resolution, because you capture more color information and can use a weaker anti-aliasing filter. See my post here.

1D is not fullframe

[Archon-X]

The 1D is not full frame.
The 1Ds, however, is.

The summary is misleading

[Anonymous Coward]

The person writing the summary mentions the imminent announcement of a 1D successor. That's wrong. The imminent announcement is for the 1Ds Mark IV. The 1D announcement was made a year ago (1D Mark IV), and it won't be replaced for 2 years.

This is an interesting demonstration of Canon's sensor capabilities. Their R&D is one of their big advantages.

Having a sensor of this resolution allows them to experiment with things like pixel binning, and using unusual pixel filters (something other than the standard Bayer matrix). I would be curious about the experiments this sensor permits.

Fascinating to read all the negative comments - I guess those grapes are really sour :)

I don't really need that many pixels...

[Anonymous Coward]

I don't need 50 megapixels, or 120 or whatever. 25 would likely do fine (about the same as film). What I want is a full frame sensor. I want an image sensors thats 35x24 mm, just like a roll of film. Its all about how many photons of light get into the camera. Fewer photons=crappier image. A billion pixel resolution on a sensor the size of the head of a pin is basically a pin hole camera, and looks worse than a crappy point and shoot. And I want it to be sensitive in low light conditions. If I can see nocturnal clouds, it should be able to image them. If I see the aurora borealis, it should be able to image it. If I set it to capture high speed events, it should capture the wings of a hummingbird in flight. And last, I want something that I can focus. If I want to take pictures of smoke, I want smoke, not the thing behind the smoke. If I see a swarm of flies, I want the picture of the swarm, not whats behind them. Is it that much to ask? Is it that expensive to get?

Re: Foveon Sensor

[EETech1]

Has anyone here used a Foveon Sensor based camera?

Are they still being made / updated?

How do they comparee to some of the cameras (sensors) described here.

Benefits / Drawbacks?

Last time i saw one was when they came out, there was a segment with Leo LaPorte on TechTV. I always thought it was a great idea. and wondered if it ever became relevent.

Thanks!

Re: Foveon Sensor

[Estanislao Mart�nez]

You may want to read this page. But basically, the way I see it, Foveon is stuck with very unpopular cameras (Sigma) that get passed over for reasons that mostly have nothing to do with the sensor tech; and because of this, there is less money invested into improving it, which means that it lags compared to advanced Bayer designs.

Bad, bad idea.

[Estanislao Mart�nez]

The alternate solution ceoyoyo is talking about requires a different kind of sensor. Imagine if you had two kinds of pixel sensors, one sensitive and the other insensitive. You'd alternate them on your sensor, perhaps in a checkerboard pattern, but basically pairing adjacent sensitive/insensitive pixels. Now, if your sensitive pixel registers too high a value, then it's probably blown out so use the value from the insensitive one (which is by definition not as bright). If the insensitive one registers too low a value, then it's probably too dark, so use the sensitive one (by definition not as dark). The crucial difference here is that you choose one over the other, and never average.

But if you do this, half of the area of your sensor at any given time is not recording light. You've effectively made your sensor perform like one with just half the area.

What about file size (100MB+)?

[markdj]

These pictures should be at least 100MB+ in size! How long will it take to save to SD card, and how big a cache is needed to do multi-fps pictures? Also how long would it take to upload these to a PC?

Re: Foveon Sensor

[hazydave]

Sigma just released another version of their Foveon-based DSLR, but it's using the same 4.6Mpixel sensor they introduced five or more years ago.
http://www.sigmaphoto.com/shop/sd15-digital-slr-camera

They're kind of selling snake oil. They make a big deal about the fact the these cameras deliver true RBG color per pixel, but then go on to advertise the number of sensor sites as the number of pixels, which is totally incorrect. They're falling up against the human eye -- we have about 120M luma-sensitive sensors per eye, but only about 6M chroma-sensitive sensors (and only about 10% of those blue-sensitive). We care a great deal about resolution, less so about color. And the resolution you get from these is 4.6Mpixel. Foveon doesn't seem to have had much of a second act, even after Sigma bought them.

There are other issues. The Foveon design is very clever, but kind of flawed. They're counting on the natural color filtering properties of silicon, and while it's a cool idea, the color filtering is somewhat off. So they have to do lots of image processing and color correction on the final result, anyway. Back when everyone else has 6-8 Mpixel sensors, you could make an argument for the Foveon chip, particularly for portrait photography, where you care more about color than resolution (well, except when shooting crusty old geezers in monochrome). But in a day when you can buy a Canon or Nikon DSLR at $700-$800 with an 18Mpixel sensor and full HD video capabilities, this one makes no sense.

As far as the cameras, the Sigma DSLRs are mighty expensive for an entry-level model with a 4.6Mpixel sensor. Sigma calls them "pro" models, but that's marketing speak for "we don't make anything higher end". Much in the same way Pabst is a "Premium" beer.

The one thing I liked from Sigma is the DP1/DP2. They had issues, but the notion of dropping a full-sized APS sensor, even the Foveon chip, into a roughly pocket-sized P&S is a good one. Of course, today, we have Sony NEX and Panasonic/Olympus doing similar sized cameras, only with interchangeable lenses.