Slashdot Mirror
What to Fight Over After Megapixels?
NewScientist has a quick look at where the digital image crowd is headed now that the megapixel wars are drawing to a close. Looks like an emphasis on low-light performance and color accuracy in addition to fun software tools are the new hotness. "For years, consumers have been sold digital cameras largely on the basis of one number - the megapixels crammed onto its image sensor. But recently an industry bigwig admitted that squeezing in ever more resolution has become meaningless. Akira Watanabe, head of Olympus' SLR planning department, said that 12 megapixels is plenty for most photography purposes and that his company will henceforth be focusing on improving color accuracy and low-light performance."
BTW, that blank before the 'm' after the sensel size was the special 'u' used for microns; Slashdot's lame filtering cut it out. Sorry.
I've fallen off your lawn, and I can't get up.
Optical SVG - the ultimate! Forget pixels. Have cameras sketch accurate SVGs of a scene with the ability to show or print at any resolution.
Good luck with that one. It's a lot harder than it sounds. Try tracing a simple 2-color bitmap in Inkscape sometime and zoom in real close. Now try tracing a full-color, full page photograph in the maximum number of colors possible.
Oh, BTW, hope you got lots of RAM and time to wait....
My blog
http://graphics.stanford.edu/papers/lfcamera/
http://www-bcs.mit.edu/people/jyawang/demos/plenoptic/plenoptic.html
1. Small phyiscal size (I wanna slip it in my pocket).
3. Good telephoto lens.
I think those two are mutually exclusive.
I want peace on earth and goodwill toward man.
We are the United States Government! We don't do that sort of thing.
What about the Red Epic 617 that will deliver 261.4 megapixels at 30fps, that's supposed to be available for $53k next spring?
I had thought that Japan's 4320p HDTV (33 megapixels) cameras were nuts, but Red's sensors are pushing far far past that.
Cameras and displays are getting to the point that they push more data than any network we've built (and so are obviously many orders of magnitude faster than the human optic nerve).
The megapixel size, like the battery life, the clock speed on a CPU, the amount of memory, etc, is mostly used in ad copy to make people think they are getting a better product. In most cases what has in fact happened is that designers put in a badly integrated laundry list of features so that even though the components sound good, it end up being a cheaply made crap product. We see this, for instance, in computer with fast processors but slow front side buses.
The point where it is going to make sense to go to a higher megapixel count is when we move to a full size 35X24mm CCD. What is happening right now is that the pixel density is getting so tight, we are not seeing appreciable quality. Additionally, I don't think the current CCDs utilize the full field of the lens. Right now cameras like the D3X is relatively expensive, but as production ramps up we may see cameras that use the full size CCD appear in the 2000 price range. At that point the 25-30 megapixel will actually be useful. The density will drop from 4000 pixels per square mm to 3000 pixels per square mm. I have seen no definitive answer on this, but some have suggested that the CCD goes as high as 4000, the noise can become a big issue.
Which is just to say, the megapixel war for the past couple years has been a gimmick, and what we might be asking for instead is larger CCD and better optics, even if the number of pixels does not change.
"She's a scientist and a lesbian. She's not going to let it slide." Orphan Black
It's fundamentally a physics issue. Light decreases at the square of the distance. The further away you are, the more light you need to gather, and so the larger your lens has to be. Sensor size doesn't matter.
If you don't collect enough light, the only way to make pictures bright is to take looooong exposures, which means you can't capture motion. (and many things you want to capture at a distance — sports shots, skittish animals — are in motion)
16*20*300*300 = 28.8 megapixels.
I agree that 12 megapixels is fine for most people who never make big enlargements (or for those who do, do not care about viewing detail in these prints from a couple feet away).
Heck, I still shoot 4x5" large format simply because the quality is amazing even in a 8x10 print. They say the eye can only resolve 400dpi or so, but my prints say otherwise. 4x5" sheet film scanned at a modest 2000dpi gives (4*5*2000*2000) 80 megapixels.
The dynamic range of our linear sensors is the weakest part of the chain. Film sucks compared to modern digital in all ways except their response curve: many films don't capture light levels in a linear way, so they can discriminate details in the clouds in a bright sky even while capturing details in the shadows. Almost all digital sensors are on the order of 9~12 stops of acceptable dynamic range, and they've been there for nearly a decade.
Cameras tend to expose for the midrange automatically. To avoid blowing the highlights, which is very visible on a screen or printout of our photos, we have to artificially adjust that exposure, called "stopping down," until we capture details in the highlights, at the expense of detail in the shadows.
There are some combinatorial techniques to achieving high dynamic range; you take multiple exposures and mathematically or artistically mix them to achieve both shadow and highlight details. But this technique is not well suited to movies or still-shots of moving scenes.
Sensors need to get a LOT better at achieving a dynamic range of 20 stops or more.
[
That is about 10^16 photons per square meter. Of you cram 10 MP on a 5x5 mm sensor, that is 3000 photons per pixel. Each pixel has a color filter that on the average transmits 25% of the photons, which means 750 photons per pixel. Simple Poisson statistics means that you get a noise that is 1/sqrt(3000) = 4% for these numbers. That is if the sensor has 100% effectivity and no electronic noise.
Avantslash: low-bandwidth mobile slashdot.
Colors are always off. No two CCD brands are color calibrated in the same way. If you want accurate color, just shoot in RAW mode, then create or obtain reasonable color profiles for the camera and all your devices, at which point it's a non-issue. If your camera can't shoot in RAW, there's your problem.
As for low light response, the easiest way to get better low light response is to use bigger optics. The light gathering of optics is directly proportional to the area of the lens (the square of the radius of the lens). The big problem we have is that camera makers are trying to use progressively smaller lenses for easy portability, and that is directly contrary to the goal of improving low light response. They have to make huge strides in response just to break even.
Until the quest to keep making cameras smaller stops, the low light performance will continue to regress. It's basically unavoidable. At best, you could improve the noise response in low light by using Peltier junctions or something to cool the chip, but there goes your battery life.
Check out my sci-fi/humor trilogy at PatriotsBooks.
Um, not quite. If you want to have the same magnification, you have to make the focal length / sensor size constant, so your position relative to the subject doesn't change, so the whole "Light decreases at the square of the distance." bit is irrelevant.
The focal length compared to the diagonal size of the senor determines what is considered "telephoto".
The aperture ratio (or f-stop) says how much light the lens gathers per unit area of sensor, and is irrelevant of sensor size, but is very dependent on focal length. (It is focal length / appearant pupil size) A shorter lens can get away with a smaller virtual entrance pupil.
So a 10mm f/2.8 lens could be very small, and given a small enough sensor that could be a good telephoto.
On the other hand, a 100mm f/2.8 is about 10 times bigger, and on a APS-C sized sensor that is a moderate telephoto.
If I have nothing to hide, don't search me
You need to learn to pre-focus the camera by pressing the shutter half way down and waiting for the right moment before clicking through. Or set the camera to infinite focus by default so that you can grab and shoot at a moments notice.
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
Just as a tip, I have found that the 50D's noise drops *dramatically* if you expose for +1 Ev (RAW) and then in post, pull back -1 to -1.5 Ev. There are lots more bits of resolution available in the stops as you move to the right (each stop has twice as many levels as the darker one to its left), and this gives the camera a one stop noise advantage over the way Canon hands it to us.
Canon seems to think we need a full stop of headroom in dynamic range over the brightest spot in the image being taken... that's simply not the case unless you're going to be compositing something brighter into the image. Seriously, try it. That +1 Ev will push noise down to an amazing degree, especially at ISO 3200 and below, where the 50D's banding issues don't rear their ugly little heads.
Of course, you can do the same with your 40D, and get even better results there, too. :)
This picture was shot at +1 Ev, ISO 100, with the 50D, and then recovered by pulling -1 Ev, effectively using the sensor 1 stop to the right. Check out the original size version ("All Sizes" button over the sample image) and look for noise in the shadows, or the sky (both the 40D and 50D are notorious for noise in blue skies... blue channel in the sensor is a weak link.)
Just be careful with your metering. If the camera isn't allowed to meter the brightest object in the portion of the scene you want to capture, you're going to clip the highlights and they will be unrecoverable.
I've fallen off your lawn, and I can't get up.
This is normal. When you double the resolution, you double it in 2 dimensions. (Height and Width) This results in a four-fold increase in data size.
But 4 megapixels to 8 megapixels isn't doubling the image size, it's doubling the number of pixels. So it is reasonable to expect the file size to double, not quadruple.
As pixel density increases, image quality decreases. These companies are cramming more pixels into the same size sensor. Bigger pixels = better quality. By the same token, I think the wars could (and maybe should?) go to sensor size eventually. There are already several full-frame (35mm film frame sized) DSLR's out there now. The image quality of any of these cameras (even ones with less pixels) will blow any ultracompact point-and-shoot camera out of the water. Bigger sensors = better quality. Bring on the sensor size wars!
m@
This is normal. When you double the resolution, you double it in 2 dimensions. (Height and Width) This results in a four-fold increase in data size.
That would be going from 4mp to 16mp. Going from 4 to 8mp should only double your image size on disk.
"We returned the General to El Salvador, or maybe Guatemala, it's difficult to tell from 10,000 feet"
That's a good point. I wasn't thinking about that. My only comment in that direction would be that perhaps the image quality settings are different and/or you were previously shooting at a lower resolution than your camera was capable of. I haven't seen the option much on the new cameras, but low/medium/high quality options used to be pretty standard as a way to save on space.
Javascript + Nintendo DSi = DSiCade
The reason a higher pixel count is always going to result in larger compressed image (assuming the same subject, lighting, etc.) even when you're well past the limits of human perception is that there is still noise in the image that must be dealt with. Image compression simply tries to remove redundant information with acceptable losses or compromises. Removing redundant information depends on predictability and detectable patterns. Higher resolution images with more detail or noise become more random from the perspective of the compression algorithm.
And before you say, "let's just work on detecting noise and allow the compressor to eliminate it." The only real way to detect noise from subject is to compare multiple exposures. This process is much too compute intensive for a handheld device that must return results in a fraction of a second and sip batteries slow enough to be useful for more than a dozen shots. Trying to do it in one pass would be brutal on images where the subject is indistinguishable from random noise (i.e. a shot of the night time sky).
My God! It's full of eval()'s.
Akira Watanabe is playing self-serving marketing games. It is not about any particular number of megapixels, it is about the size of the light well of each pixel, or "pixel density" if you will. That determines the dynamic range and sharpness per pixel. Mr. Watanabe's company has pushed that past their main competitors despite the fact that those competitors have larger number of megapixels in their top cameras. How is it possible? Sensor size. Let's look at some of the offerings today on the market (the smaller the density the better):
So yes, Olympus is now hitting 5+ MP/cm2 on their SLRs despite having much lower number of pixels than Canon with their 5DMkII, Sony A900, or Nikon D3X. This is because Olympus boxed themselves into a smaller sensor (4/3). All the posturing about how 12MP is enough is only designed to hide their shortsightedness.
(BTW, /. seriously needs to allow <pre> tags)
End anonymous moderation and posting on
Poor light performance has ALWAYS been the biggest problem I've had with digital cameras. What good is a million megapixels when you can't even see your subject without shooting in direct sunlight? Low light performance has always lagged behind on digitals (most of them I've bought over the years have had the light performance of equivalent of about 200 ISO film).
SJW: Someone who has run out of real oppression, and has to fake it.
Olympus, on their 4/3 system (along with Panasonic), uses a smaller sensor than even a standard DSLR, much less a full frame. So, 12 mp might well be a physical limitation of their design, both in terms of pixel density and optic size.
I think that is where we are running into the limits of physics: pixel density, not pixel count. Of course a full frame sensor is going to be able to reach higher resolutions, but for the size of camera that consumers are interested in, the megapixel density race may well be over.
Look at Nikon, who made big waves in the recent past by releasing the D3 for $3500 with only 12 megapixels. The thing is, the D3 packs a sensor that can push all the way to ISO 25,600 with little more noise than a normal camera at 1600. That means 16x more light captured, if I'm not mistaken. I think this camera offers much more utility for the average photographer than, say, the Sony A900 with over 20 MP. Honestly, having some acquaintance with Sony optics, half of their current lenses can't begin to resolve that kind of detail.
In order to get more pixels from this point onwards, I think we're going to see folks using full frame cameras, or even medium format. Pentax just announced their re-entry to the medium format market with a ~45 megapixel camera, along with their huge range of optics of commensurate capability. Of course, this has a price, as the camera pretty much mandates a tripod, and will need something like a small suitcase to transport.
The physics tells you are eventually limited by the point-spread function of the lens system.
This is more or less where we are at right now. Any more megapixels will simply mean extra blurry megapixels.
But putting anything above 10 megapixels in a tiny form factor is a total waste. I shoot with a Nikon D2x and D300 - both 12 megapixel cameras. With these cameras and decent lenses you need excellent shooting skills to really tax the quality of the file. That means fast shutter speeds and decent glass. Neither of which are common on the low end cameras. Even with a fairly fast shutter speed, say 1/200 with a 50 mm lens, you can often see visible differences between a hand held shot and one set up on a tripod. There is that much data in a 12 megapixel file. Of course, that is pixel peeping which most point and shoot camera users won't do.
Big files slow the camera down - more data to push through. Slower overall response. More battery used. Bigger files, harder to email. Not saying that the manufacturers won't try this, but it's pretty pointless. There are lots of other ways to improve picture quality - better high speed capability, stabilized lenses, better optics.
Faster! Faster! Faster would be better!
No, it is misunderstanding physics.
The more pixel sensors there are, the more space there will be between them relative to the total area. Increase the number of pixel sensors, and you reduce the area that's covered by sensors.
Take star photography as an example -- if the light from a star hits between two pixel sensors, it won't be registered. It's immensely better that it registers as a faint pixel with less accuracy than it not being registered at all.
Another example is low-light photography, where the base ISO decrease you get by increasing the pixel count leads to more artefacts and a worse result.
To always get a benefit from an increased amount of pixels here, you have to make sure that the amount of border between pixel sensors does NOT increase. But with current technology, it does.
This is, by the way, why I much prefer a Nikon D40 over a D40x. The cameras are near identical, except for the sensor, where the D40 has a 3008x2000 ISO-200 sensor, and the D40x has a 3872x2592 ISO-100 sensor. The D40 is the better camera to use for things like star photography or low light conditions, because a larger part of the D40x sensor is dead space.
JPEG compression is JPEG compression and RAW data is RAW data.
not really, there is actually lossless compression of jpeg images that will reduce size by 25% (sometimes more) from what a camera produces (not even including the removal of exif, and thumbnail removal)
You miss the point of RAW. RAW has more information than JPEG, because RAW is essentially direct dump from sensor + actual image settings. Sensors are not precisely RGB. But dumbly converting them to JPEG or PNG, you loose lots of information.
Classical example is balance of white. In JPEG/PNG it is all already pixels. In RAW you can change WB without loss of quality. Or extra bits of color, which can be used for EV (exposure value) adjustments.
It would depend on how much memory/CPU time/power the camera is willing to use. I find I get a much much better, and similar sized image post processing on my PC, than selecting at the camera. (IE minimal jpeg compression at the camera produces a larger, and noticeably worse image, than doing the downsize with imagemagic, from a full size RAW format on a PC)
ImageMagic isn't a RAW development tool.
If you want something on cheap, try RawTherapee or GIMP+dcraw (on Linux; GIMP + UFRAW on Windows). With GIMP/dcraw, you would be able to adjust RAW parameters ("develop a RAW") during import and then in GIMP do whatever you like with result of development.
All hope abandon ye who enter here.
If you can afford to have a team shooting an event, then you can afford Aperture, Lightroom, or one of several other products that lets you do automated renaming based on, amongst other attributes, Camera Serial Number. As long as your shooters aren't swapping cameras with each other, there you have it.