Slashdot Mirror
The Future of Digital Camera Technology
An anonymous reader writes "CNet News has an interesting look at where digital camera technology is headed now that the megapixel buzzword can be put to rest. From the article: 'In compact cameras, I think that the megapixel race is pretty much over,' says Chuck Westfall, director of media for Canon's camera marketing group. 'Seven- and eight-megapixel cameras seem to be more than adequate. We can easily go up to a 13-by-19 print and see very, very clear detail.'"
Now that we have cameras of a decent MP maybe we could stop saving as jpeg and instead use a lossless format?
Er, it's called RAW, and all pro cameras and even a lot of pocket cams are capable of using it. This is unrasterized data, so it's about as lossless as it gets (even TIFF is destructive because it rasterizes the images before it's saved).
The problem with RAW formats right now is that they're all proprietary, but this isn't really that big of a deal in practice. Generally speaking, if an image editor supports RAW at all, it will support every major camera. And every camera that supports RAW also ships with its own conversion software (so you can save as whatever format you want).
A little less than a year ago, a graduate student at Stanford gave a talk on light field photography at the University of Washington. The results were extremely impressive. Basically, by inserting an array of microlenses in front of the CCD, you can determine the direction of every ray coming into the camera. You lose resolution, but who needs 8 megapixels anyway? What you DO get is the ability to refocus the image in software, and take photos in low light and still retain a high depth of field.
I highly encourage you to check out his light field photography site, including his galleries, tech reports, and papers. It'll blow you away.
A 13MP RAW image is NOT 39MB. Each 'pixel' in a digital camera only has one color (red, green, or blue typically, sometimes white (Sony), other colors could be used) - it's 8, 10, or 12 bits of data (you won't find 16 bit D/A conversion in a digital camera - it isn't practical and it's well beyond the human eye's ability to discern anyway - though you could argue it would be useful for making very large corrections in saturation or brightness without losing quality) - so stored with no compression at all this is at worst 13 x 12bits = 21.5mb. Add in the fact that you can get a decent compression ration across this data (and your typical 6-8MP DSLRS certainly do) without any loss of data ... maybe 15mb ... or less.
Want to save an image losslessly? Digital SLRs (and some point in shoots) let you save in RAW. Not only is the image saved losslessly, but you can adjust white balance, exposure (within reason), sharpness, and more all after taking the photo!
Want decent optical zooms? SLR lenseshave been available for decades now that range from 8mm to a whopping 1200mm. That's over 100x for you guys used to talking about lenses in terms of "how much zoom" they have. Canon's lens selection.
Want 13MP? The Canon 5D does 12.8. The Canon 1Ds Mark II does 16.8.
Want storage? You can get CF/SD cards as large as 8 gb, and portable hard drives such as the Epson P-2000 made for offloading photos out in the field.
Want to wirelessly transmit photos? The 1Ds Mark II can do it with the WFT-E1A.
Current batteries can let you shoot 2500 shots on a single charge. Spare batteries are cheap and keeping spares in your bag is no big deal.
The thing is that all this technology is already available, but be prepared to spend thousands of dollars for it. If you're looking for all this technology crammed into an everyday point & shoot, give it a few more years.
Instead of more MP, how about better high ISO capabilities? No shutter lag from when you press the button to when the camera takes the picture? How about taking photos at 8fps? Instant-on when you power up your camera? Quicker autofocus? These features are very important, but these too are available on DSLRs, and for a price. Considering how little money you're spending on a point & shoot, they do quite a bit as it is and they'll only get better. The technology is already there and it will eventually find its way down to lower end cameras.
"They tell you the camera has 8MP, but "forget" to mention that in reality it has 4M green pixels and 2M of each red and blue. And there's a blurring filter in front of the sensor to reduce moire. So if you photograph fall foliage, your 8MP camera turns into a 2MP one at best. In the BEST case, it's a 4MP camera really, not 8MP.
The only sensor that takes full RGB readings at each sensor location is Foveon, but it suffers from inferior color reproduction and lower ISO sensitivity. It's also pretty low on "real" pixel count - currently at around 3.5MP (which in Canon/Nikon terminology would be called 10MP, because each pixel takes full RGB readout). Foveon pictures are extremely sharp, though, and render textures very well."
This utterly fails to take into account how the human visual system works. It also fails to take into account the necessity of filtering when sampling. It also fails to take into account the sophistication of current interpolation algorithms.
The Bayer pattern is actually just about the most efficient layout for capturing images for human perception. I have done dozens of camparison of images capture using the 6Million Bayer arrayed sensors, versus 10.2 Million layered sensors. In the end they are essentially equivalent. The bayer layout allows you to do more with less by taking into account the human image processing system that is heavily organized to toward luminance/green information.
It is utter fanboy nonsense to say a bayer 8MP camera turns into a 2MP when taking fall foliage shots. In any real world situation including fall foliage, an 8MP bayer camera like the Canon 350D will capture more detail than the Foveon sensored SD10 NEW 10.2 Million Pixels (3.4 Mp Red + 3.4 MP Green + 3.4 Mp Blue) (description from Sigma USA page).
As technical bunch we should be able to understand that optimization is sometimes better than brute force. By tilting the sensor toward green, it is tilted toward luminance capture and tilted toward the way humans view details.
In thousand of empirical comparison online, parity is reached when there is an approximately equal number of green sensors. So 6MP bayer (3MP green) where approximate equal to 10.2MP foveon chip with ~3MP green. Actual 10MP bayer (5MP green) cameras like Nikon D200 easily capture much more detail than Sigmas 10.2MP chip.
The sampling issue. The Sigma has no filter to prevent undersampling artifacts. It doesn't suffer from colour moire artifacts, but it has plenty of luminance moire. See here for an ancient comparison of the 6MP Canon D60 and the 10.2MP Sigma SD9:
http://www.wfu.edu/~matthews/misc/DigPhotog/alias/
Scroll to the photo comparison at the end. The only extra detail in the Foveon based image is Aliasing errors. These are extremely prevalent in Sigma images with sharp diagnals, or repeating patterns beyond the Nyquist frequency of the sensor.
In the end, bayer is an excellent engineering optimization to do more with less. The real comparison that counts is how does it compare with film. A 6mp Bayer sensor in an DSLR is already better than 35mm film. By 10MP it is significantly better.
The other important factor is how the bayer DPI translates in the printed image. I have found that around 240 DPI is close to optimal image quality. So a Canon 350D with a 3456 pixel image width can produce a superb quality image about 14 inches wide. Be aware this is not to say you can't print larger. This is highly subjective depending on source material, but with detailed material this is the point where I consider that you would be hard pressed to notice any improvement from more pixels.
So even if you only want to print 13"x19" I think you could still see improvement from more pixels if printing detailed subjects like landscapes.
You can argue the quandry of subject, material and view distance till the cows come when considering viable prints size. I mere wish to express what I consider the