Slashdot Mirror
Improving Digital Photography
Milican writes "'It's easy to have a complicated idea," Carver Mead used to tell his students at Caltech. "It's very, very hard to have a simple idea...And now one of Mead's simplest ideas--a digital camera should see color the way the human eye does--is poised to change everything about photography. Its first embodiment is a sensor - called the X3 - that produces images as good as or better than what can be achieved with film.'" We had a previous story about Foveon last February.
How is this at all like the way the human eye sees?
I hate pixel noise in my digital pictures. I have heard that since red color has to be detected at the deepest part of the silicon there is an abudance of noise in the reds.
0xfeedface
wait till a few years down the road once he's up to X10!
In SOVIET RUSSIA... erm...NSA AMERICA, the Internet logs onto YOU!
in Photography. Check out the article here.
Random is the New Order.
for an excellent (as usual) review of a camera based on this sensor check dpreview
http://www.dpreview.com/reviews/sigmasd9/
-- the cake is a lie
"It's very, very hard to have a simple idea."
I don't know about anyone else, but this GW Bush bashing is getting a little tiresome.
Best Windows Freeware
Before all of the replies saying that digital is for geeks and film will forever rule, please be sure that you have used current and professional quality digital gear, including 35mm gear made by Canon or Nikon with standard lens mounts, digital medium or digital large format backs (depending on the type of vs. film comparison you plan to make).
Consumer digital cameras are one thing... X3 is another (still hotly debated)... but most photo editors and labs out there right out agree that a Canon EOS-1D, EOS-D60, a Fuji S2 or a Nikon D1X or D100 is simply takes better pictures in nearly every regard (including resolution) than a 35mm film camera, with any brand or grade of film. With the latest range of full-frame cameras such as Canon's EOS-1Ds (11 megapixel, I believe) and Kodak's 14 megapixel offering, the distance between digital and film (with digital on top) will only increase.
And before you comment on other film sizes, realize also that many of the largest advertising companies shooting commercial spreads abandoned film long ago and are shooting with digital medium format or large format backs. Yes, many of the fashion or product spreads you see in your favorite checkout stand magazine are in fact digital these days.
Film is well on its way to becoming a playing for history hobbyists and an art tool for retro artists, and no amount of "ludditing" will change this.
STOP . AMERICA . NOW
I also think it should be noted that this Popular Science article (and the submitted article) make it clear that it wasn't Carver Mead who invented it/thought of it but Dick Merrill who thought of it and Dick Lyon who brought the dream back to life after Merrill forgot about it. Mead just founded Foveon Inc.
Random is the New Order.
It sees a real "color" instead of on red/green/blue (dispersed in fine pixels of course). It may not be able to see red quite as well as other colors, but it only means that the sensitivity at the red level is the limitation you have for the picture as whole.
What you don't get is Moire patterns - at all!! That is what you probably hate when you say you hate "pixel noise" because it's totally obvious (due to the color changes), very distracting, and annoying to clean up after.
"There is more worth loving than we have strength to love." - Brian Jay Stanley
A good review is at dpreview.com (skip to conclusion if you're in a hurry).
This technology still has a way to go, but the SD9 certainly is an interesting camera.
One huge problem is with adaptation - Sigma makes consumer-grade lenses and cameras, some of which are of poor quality (but quite affordable). For these cameras to be adapted by professionals, Sigma need to create a camera with Canon or Nikon mounts, but furthermore, they need to erase the stigma attached to their equipment by many professional photographers.
If they were to make a full-frame sensor in a Canon mount that worked better at higher ISOs, this camera would be a huge seller.
Since this new chip is able to gather more light than traditional CCD chips, I would imagine that there will be some interesting uses for it in astrophotography. Instead of having to use a CCD imager with a 30 minute exposure to get an image, wouldn't you technically be able to get a higher resolution pic with this a lot quicker?
That's just a thought...
The minilab system that is widely regarded as the best is the Fuji Frontier system. How does it work? By scanning film. Of course, it accepts files from digital cameras as well.
What is the best way to get large, "professional" prints? The Lightjet. How do these operate? Using very high quality scans! (See West Coast Imaging, for example). My point? You can already get digital images produced in the exact same manner as the best film prints.
There are already a lot of people who think digital photography has surpassed even medium format photography. See the Luminous Landscape, for example.
As for widespread adoption, photojournalists have all but abandoned film. The P&S crowd is already beginning to abandon film.
Great, now I can stop scanning in those 21Mpixel images from film, and get a 10Mpixel digital camera. Since it uses 3 layers, those pixels must count for more than twice as many from the 35mm film. And the dynamic range is surely greater tha slide film. Finally the shadow detail in that otherwise brightly lit scene that I needed to use slide film, and capture at 48bit can be resolved with a 24bit image! Now I won't need more memory - my files will be 1/4th the size, and look just as good!
And it sees just like we do! Same 3 colors, same intensity relations, all on each pixel! Because everyone knows the human eye has only one kind of sensor in it. It's not like mammal eyes that have rods and cones.
Sorry, film will be around a little longer....
- dave f.
No new screens would be needed. This new sensor only affects the way an image is captured, not how it is displayed. Current CCD chips actually use 4 "pixels" to record each pixel of the image. 1 red sensing pixel, 1 blue sensing pixel, and 2 green sensing pixels. It is set up like the following for each pixel the camera records...
:)
RG
GB
The CCD device in a digital camera has one of these set up for every pixel the camera is to capture.
This new way will allow all 3 colors to be captured on one "pixel" instead of 4, so that will allow much higher resolution pictures to be taken. Hopefully this simplified explanation makes sense, and didn't totally confuse everyone
I hate to break it to y'all, but in the human eye, each spot in the fovea is occupied by one receptor, which is maximally sensitive at one wavelength -- in other words, it works the way that current digital cameras work. (Random Googled link.) I suppose that if the human eye needed to determine the color of a particular "pixel", it would have to interpolate, just like a CCD camera... but that's a moot point, because that doesn't actually happen in our visual system. (It's much, much more complicated than that.)
Now, this technology does sound like a great way to increase the resolution of digital cameras, if it's feasible. However, all this "neuromorphic" stuff is pure marketing. (Though I admit that "Foveon" is a clever name.)
MSK
t's pixelated still so you will still get Moire patterns as soon as the smallest details are finer than the resolving power of the X3 bins (think Nyquists theorem). However, the bizarre colours you get from a fine-grained black and white grid shouldn't be present to the same extent as all the measurements of colour intensity are done at the same point in the X3 layer, as opposed to the different spatial positions of the red green and blue bins in a colour CCD.
The bizzare colors (what I really hate about digital photos) are not just reduced - they are gone. If you read the review at DPReview.com you'll find that it has resolution right up to Nyquist is noise free and you get some detail beyond. Here's the relevant section (near the very end of the review, where they test against some resolution charts):
The SD9 is capable of delivering all nine individual lines of the horizontal or vertical resolution bars up to its maximum absolute resolution (sensor vertical pixel count) and slightly beyond. Note also that because the X3 sensor doesn't need a color filter array it doesn't suffer from color moiré.. Absolute resolution is just less than the Canon EOS-D60, Nikon D100 and Fujifilm S2 Pro (at 6 mp).
However, because the X3 sensor doesn't use a low pass (anti-alias) filter it is able to resolve detail all the way up to Nyquist. Beyond Nyquist the system will alias without any objectionable color moiré. Where a Bayer sensor camera would turn detail beyond Nyquist (such as distant grass texture) into a single plane of blurred color the SD9 will continue to reproduce some individual pixel detail (without color moiré).
"There is more worth loving than we have strength to love." - Brian Jay Stanley
The X3 announcement came out almost a year ago, and still their is only one, ONE camera that has this technology. If its so superior (which is it by the way!) then why the hell hasn't this thing been flooding the market? It defies description.
In fact, earlier this year the announcment was that we should see several cameras with X3 technology on the store shelves in time for Christmas. What happened?
Planet P Blog - Liberty with Technology.
www.enthea.org
Film still rules for taking pictures in low-light.
So that's why the shuttle keeps visiting the Hubble Space Telescope, to pick up the film!
The is also a company called SBIG that makes a line of digital imagers for amatuer astronomers.
Steve M
As much as Foveon's well hyped and widely advertised (*cough*thanksslashdot*cough*) idea seems to make sense on the surface, their solution is far from perfect.
To sense an RGB (Red, Green, Blue) pixel one can use a veriety of methods. At the center of this technology lies the ability to turn a stream of photons into an electric current. This photodetector is colorblind, it is only capable of measuring the _amount_ of light, not it's color. To recognize color the estheblished method used to be to put several photodetectors near each other and put color filters in front of them. The most widely used color filter array is known as the Bayer pattern and consists of 2 green photodetectors (diagonal from each other) a blue and a red detector in a 2x2 grid. These 2x2 blocks are then repeated over and over to create the full image sensor.
Specialized software or hardware needs to take these individual Red, Green or Blue pixels and recreate a single RGB pixel, this technique is known as demosaicing. The major advantage of this method is the simplicity of the photodiode (photodetector). It allows for the creation of very dense image sensors that are now passing the 10MegaPixel barrier while keeping the cost down (start seeing 5MegaPix sensors for less then $100 before the end of this year).
Foveon's approach is to layer these color filters vertically.
The good:
- idealy you get R,G,B at each pixel.
The bad:
- very complex layered photodiode technology, this makes the pixels significantly bigger. Currently the pixels are bigger then a 2x2 bayer image pixel. The complexity also adds to the manifacturing cost, these chips will not be cheap for the forseable future.
- Color bleeding. For example: Photons in the green wavelenght do not nescecarily stop in the green layer, but might be picked up by the underlying red layer. This means that specialized hardware needs to apply a non-trivial color correction for each pixel layer.
Foveon's idea is a very interesting approach. Since they nicely pattented their idea shut, we will have to patiently wait for this single company to provide the world with this technology.
Side fact: The human eye see's colors using pigments that respond differently to different wavelengths. In the simplest model we can say that we see Red Green and Blue with spatially seperated pigments that resemble a bayer image sensor closer then the foveon's sensor.
There is no way this camera sees like the human eye - this sensor arrangement is completely different from the rod/cone structure of the human eye. A conventional digicam is actually closer than this is.
As far as this camera comparing it to film - more baloney. A good 35 mm camera on a tripod is capable of somewhere 11-14 megapixels of in a conventional digital camera. This particular sensor does not deliver resolution in that ballpark.
Too much hype. All they did was stack pixel detectors rather than mosaic them. The mosaic was simpler and now cheaper, this thing costs $1800 in a camera, else I'm sure someone could've come up with it. The real accomplishment is creating those silicon layers precisely, not coming up with lets stack em
They say the resolution is like a 120mm film, and the color lattitude is big. So are CMOS sensors in Canon and Nikon's cameras. Checkout the awesome photos on photo.net. A lot of those have been shot by modern digital cameras with CCDs and they dont look bad. Mead has his own marketing to do to try and take Foveon to Intel and Microsofts level, so he has to push down CCD. Theres a reason why people are buying digital cameras with sensors smaller than fingernails and submitting their pictures on professional photography site. I think Mead has work to do.
"Give orange me give eat orange me eat orange give me eat orange give me you." -Nim Chimpsky
It's a neat technique to increase resolution, but the implication that the article gives that you need this technique to improve resolution is silly. Effectively each grouping of red, green, and blue sensing points in a CCD camera returns a single pixel. If you replace each red sensor with three smaller sensors (one red, one green, and one blue), you'll get the same increase in resolution. In theory you could lose data because a little bit of blue light hit the red sensor, but not the blue one, but in practice it isn't an issue. Assuming you can keep making the sensors small, you can keep scaling the resolution of CCD technology.
This is neat technology and may well improve the quality of cameras to come. But it's not essential to improving the quality of cameras.
Search 2010 Gen Con events
The resolution (as determined by number of pixels) will not get better. Manufacturers are currently counting each one-color pixel in the
RG
GB
blocks as one. That block is 4 pixels. Foveon-based cameras would have
(RGB) (RGB)
(RGB) (RGB)
which is still 4 pixels, but gives you more accurate color information at each pixel and reduces moire. So, while there will not be any more pixels per area with Foveon CCDs, the *effective* picture resolution will be much better.
I wish I had known this before I shopped for digicams-- it feels like false advertising to me, and I learned after I had made my purchase. Manufacturers ought to be required to state "4 single-color Megapixels" or "1 Megapixel effective with color" for 4MP cameras with traditional CCDs.
This is amazing technology, and it will revolutionize digital cameras if/when it comes down in price. HOWEVER, this is not how the human optic system works. Even in our optics, we have seperate receptors for red, green, and blue, and our brains do the interpolating. As most will remember from basic elementary biology, our eyes detect light through rods and cones. All quotes are from this link. "The retina has ~126 million photo receptors, 120 million rods and 6 million cones." Rods gather any light they can, and compile the data together to show the best possible image in the dimmest light; therefore, rods will display a black and white image. This is why the darker it gets, the harder it is to differentiate yellow from white: you are depending more and more on the rods.
HERE is where it gets interesting, and where I get to my point. Cones are what we use to see color. An individual cone cannot see red green and blue as this marketing hype would lead us to believe. "The cones come in three types: Red (60%), Green (30%) and Blue (10%). The red and green cones are randomly distributed in the center of the fovea and the blue cones form an annulus around the outside." So in effect this camera will actually surpass the human eye.
As a side note, the link goes to a very interesting document that states how "126 million photoreceptors must be transmitted to the brain via 1 million fibers in the optic nerve [while] [t]he overall compression ratio of 126:1 is not evenly distributed." Check it out.
I just finished reading the review at dpreview. (Thanks to all the people who posted the link). There may be a serious issue with this technology. In the review they mention "color clipping". Once one of the color channels reaches saturation, all color information is lost. This may be inherent in the X3 design.
The detector works by the difference in absorption of the colors of light. The first layer sees a lot of blue, with some green and red. The next layer sees a lot of green with some red and a little blue. The last layer sees a lot of red with only a little blue and green. What this means is that in order to determine the true colors of the reverse of this process needs to be calculated. However, if any of the detectors saturate (and the first is the most likely one), there probably is no accurate way to do this reversal. Currently, it looks like the camera makes these pixels grey, which looks aweful. They will need to come up with a better way of estimating the color of these pixels if this technology is to work well, and I have no idea if that's possible.
Note that a standard CCD with separate pixels can also have one of it's channels saturate. In this case, however, the pixel will simply become whiter than it should, which looks natural.
Basic PNG can store images with up to 48 bits of depth without a problem, and the basic compression algorithm is what's used in gz - it's deliberately patent-unencumbered.
Also, the statements of some slimy money-grubbers to the contrary, the jpeg compression scheme is patent-unencumbered as well, and the JNG format (one of the PNG family) allows 12 bits per channel per pixel.
See the technical specs on libpng.org for more details.
That is a good point that noise is error in a way that grain is not... I would like the SD9 to have noise levels like the S2 pro (which is great in low light) but given what I like to shoot (landscapes and architecture mostly) I will be happy enough with what the SD-9 can do. One of the things I like about noise on the X3 is that at least the noise doesn't result in distracting colors like you get when individual CCD sensors have a different level of noise... I'm not sure if each "layer" of the X3 sensors have similar noise character (where each layer has its own problems) or if they are more in line with each other, resulting in less color loss from noise.
The artifacts you talk about worry me a little too, I have been reading the Sigma SLR forum on DPreview now for a while and I have seen some samples with color fringing and washed out greens. The green problem in particular seems to be an overexposure thing and so I think can be managed with careful exposre. I'm not sure what is going on with the fringing, hopefully that will improve as they fine-tune the firmware. Also, I think there is some open question around what the Sigma software itself is doing with the raw data - the open source raw converter is handy in that you can take a raw sigma image, turn it into a PPM, and examine it from there. I've been starting to do that to make the final descision about this camera.
I do think having to have grain in a photo to accept it is something that will change over time as people become less and less used to seeing images with grain. It seems like a lot of professional images you see around now are very smooth with nothing like grain or noise, but perhaps even these have subtle amounts I haven't noticed, and you get texture of soome sort from some printing processes.
"There is more worth loving than we have strength to love." - Brian Jay Stanley