Full Color Capture: Hype or Hero?

Background

You may have heard about the upcoming Sigma SD14 that offers full color capture, but do you know what full capture is and what it can do for your photos? Will the full color capture SD14 set a new standard for digital cameras or will it be a mere curiosity like it's older siblings the SD9 and SD10 which developed a loyal following but never quite turned the tables on sensor design as originally hoped. As of this writing, the Sigma SD14 is not yet out, but the technology is already in place so let's take a look at the technical details of full color capture versus single color capture.

Single Color Capture

The vast majority of digital cameras including high end professional dSLR's use an image capture sensor that can record only one color per pixel. Most sensors use what is often referred to as a "Bayer mosaic" pattern where the sensor only records one of the three primary colors (red, green, or blue) at each photo site (pixel). A six megapixel dSLR, for example, may have a sensor with 3000 x 2000 resolution. One thing that is often overlooked is the fact that each of those "pixels" on the sensor only records a single color: red, green, or blue. To make matters even more complicated, single color capture sensors do not divide their pixels evenly, recording 1/3 red, 1/3 green, and 1/3 blue. Instead, half of the pixels on the sensor are green while only 1/4 are red and 1/4 are blue. More green sensors are used because having greater sensitivity/resolution in green mimics how the human eye captures color. The RGBG layout of a standard digital camera sensor looks something like this:

Since a 3000 x 2000 (six megapixel) dSLR returns a full color image with all three colors present at each pixel, the most obvious question at this point is how we can end up with a full color image when only one color was recorded for each pixel on the sensor! The answer lies in interpolation. Digital cameras and raw processing software use sophisticated algorithms to predict the missing two colors at each photo site (pixel). As an example, take a look at a blue photo site somewhere in the middle of the above graphic. Notice that at every blue photo site, there are four red photo sites adjacent (diagonally) to the blue photo site. If all four of those adjacent red photo sites have high red brightness, it can be "assumed" that the blue pixel will also have high red brightness. This is a simple example but similar prediction-based algorithms are used at all other pixels to recover the two missing primary colors for each pixel until each pixel has all three colors (one actual, and two predicted). Obviously the algorithms get much more complicated when surrounding photo sites are not the same brightness, but the general idea is to "guess" the missing two primary colors at any given pixel by looking at the color of surrounding pixels. Once both of the missing primaries have been interpolated for each pixel, the final full color image has been reconstructed.

Problems with single color capture

The above single capture Bayer Mosaic sensor is used in nearly all digital cameras as of this writing. If you are familiar with interpolation, you probably already know that interpolation comes with certain drawbacks. Because a single color capture sensor only captures one of the three needed colors at each photo site, two thirds of the information in your photos is being "guessed" while only one third is "real" data! By the numbers, you'd have to wonder how this even works at all! The answer lies in the fact that our eyes are more sensitive to changes in detail, edges, and brightness than changes in color. In addition, the interpolation algorithms used to reconstruct the missing colors at each pixel have become so advanced that they actually do a very good job at predicting the missing colors under most circumstances.

The real issue with single color capture sensors arises when you have subjects that have colors close to the primary red, green, and blue colors used for the photo sites on the sensor. For areas of detail that are black/white, all photo sites on the sensor will be reacting similarly (will have similar brightness). This makes it easier for the interpolation algorithm to reconstruct the image because each photo site will be recording near the same values. This is why, when reviewers shoot resolution charts, the cameras return resolution numbers comparable to what you'd expect if the sensor were actually a full color capture sensor recording all three primary colors at each photo site.

When the balance of color starts to shift however, particularly toward red or blue, things start to go downhill. When shooting a bright red flower with dark red veins that only "excites" the red photo sites on the sensor for example, you can see by the graphic above that your resolving power quickly drops to near 1/4 resolution. This is because the green and blue sensors simply offer no data (they are black) and only the red sensors contribute data. The same would be true of a bright blue sweater or blue fabric. While black/white subjects may be resolved at near full resolution, some red/blue subjects may fall to near 1/4 resolution and other colors like yellow, green, orange, etc. fall somewhere in between. Of course, you don't see this difference as missing pixels: only a loss of detail/sharpness. The result is that you end up with an inconsistency in sharpness in photos that makes some colors less sharp/detailed than other colors, and the visual result is a bit "flatter" look that some would see as less three dimensional.

The only saving grace for the single color capture sensor is the fact that it is often difficult to find a subject that has a color so closely matched to the red, green, or blue filters on the sensor that the other two primaries receive no data whatsoever. As an example, the red photo sites on the sensor will certainly be affected more than the green and blue sites, but most shades of red will still invoke some type of response from the green and blue sensors. It is rare to find a shade that matches so well that the sensor records no information whatsoever at the green/blue sites. Granted, the lower the brightness recorded on the green/blue photo sites, the lower detail you'll have to work with for that red subject and (potentially) the higher the image noise levels.

For more information on "sharpness equalization" as a means for correcting loss of sharpness/detail in single capture sensors, please read my article at Digital Outback Photo or try the "sharpness equalizer" in my Qimage software.

Full color capture and what it can do for us

Released in 2003, the Sigma SD9 was the first camera to offer full color capture. The sensor, manufactured by Foveon, was touted to be the next generation in digital camera sensors. Using three sensor "layers", the SD9 (and soon-to-follow SD10) offered the ability to capture all three primary colors (red, green, and blue) at each photo site on the sensor. Since no interpolation was necessary, the typical problem with sharpness/detail consistency across different colors was solved and to most people, the result was a more 3D feel to images. The new technology didn't come without problems though...

The first problem faced in mass marketing this new technology was that, while the SD9 and SD10 were marketed as 10 megapixel cameras, the final images were "only" a little over 3 megapixels. The Sigmas were competing with 6 megapixel dSLRs that, to the "unwashed" appeared to have twice the resolution even though the full color capture Sigma was actually capturing more data, and doing it in a more sensible fashion. Because many reviewers base resolving power on test shots of a black/white resolution target, the Sigma performed poorly compared to the single color capture 6 megapixel dSLR competition because black/white detail is handled nicely on standard cameras. Had those resolution test shots been black/red or black/blue instead of black/white, it would have been a different story.

It didn't help matters that you can't stop the age-old rule of thumb that you need 300 PPI of detail to get a good print. The die hard 300 PPI camp would argue that they could print bigger prints using a standard single color 6 megapixel dSLR because the final image was 6 megapixels compared to the 3.4 megapixels recorded by the full color capture SD9/SD10. It also didn't help that the SD9/SD10 could only shoot in raw format and pictures had to be developed after-the-fact and that the camera body wasn't the best on the market at the time and being a Sigma body, it needed Sigma lenses which gave Nikon and Canon followers pause.

The final tether that kept full color capture from reaching escape velocity in the SD9/SD10 is the fact that it did have some problems recording consistent, noise free color. People familiar with the camera and raw developing software could produce some gorgeous photos but it did, on average, take a little more work than standard single color capture dSLRs. It turns out that the layers used in the Foveon full capture sensor made it more difficult to get consistent/accurate color fidelity compared to the arguably simpler design of the single color capture sensor. The result was that the full color capture Foveon based SD9/SD10 were a little harder to keep under control with respect to color accuracy and they suffered from a bit of metamerism (colors shifting under different light sources) that was not accounted for by the hardware/software.

Looking for a bottom line: is full color the future?

Right now, the SD14 appears to be the new contender in the next attempt to get full color capture into the mainstream of digital photography. The camera has not yet been released, but you can find information about it here. At first glance, the SD14 seems to step into the ring with some of the same handicaps that held back its older siblings. While it will be advertised as 14 megapixels because it records three colors at each photo site, it will return final (non-interpolated) images that are under 5 megapixels, less than half the final resolution being returned by the single color capture competition.

It remains to be seen if Foveon has improved color fidelity of the full color capture chip and if Sigma have made improvements to the body, but at least the SD14 is capable of returning developed (JPEG for example) photos and doesn't require raw developing tools. While I always shoot in raw mode by choice, some jobs actually require shooting finished images for the sake of time and I'm sure the ability to shoot in a "finished form" will improve sales. Final price still has not been set to my knowledge, so I'm sure that will be a factor as well.

Technically, the SD14 is an interesting camera and I applaud Sigma/Foveon for keeping the concept alive! It really has potential as it does correct some image quality flaws inherent to single color capture devices. In this respect, the SD14 is an important entry in the world of dSLR cameras! Mathematically speaking, the SD14 will record 40% more "real" data than a 10 megapixel dSLR even though the final images will have half the pixels. It sounds confusing at first, until you realize that the SD14 is investing the data in color capture rather than added pixels. Whether or not the "masses" will recognize that extra data as a benefit or a detriment remains to be seen, but if it didn't happen the first time (with the SD9/SD10), I have my doubts this time around.

Summary: The future of full capture

Full color capture resolves a number of issues related to today's single color capture sensors. Single color capture has been around for decades, however, and the sensors and the interpolation algorithms that make them work have been refined over time. Many of the pitfalls of single color capture can be addressed with advanced color interpolation algorithms. As a result, to really get noticed, I believe full color capture has to take a leap forward that would make it a clear winner in the eyes of the consumer. In my opinion, to do that, the final image resolution needs to be comparable to today's dSLRs. Regardless of how good you are with math, some will see the SD14 as a 4.6 megapixel camera competing in a 10+ megapixel market. Even if you grant that the SD14 actually records 1.4 times the amount of data compared to a typical 10 megapixel dSLR, 12-14 megapixel dSLRs are on the horizon that will match the amount of data recorded by the SD14. Anyone familiar with digital sampling and integration will realize that if you make the pixels small enough and abundant enough, it won't matter that you can't record all colors at once. Case in point: inkjet printers, audio CD's, DVD's, etc. At some point, when the pixels get small enough, it won't matter whether they are on top of each other or not!

Due to the consumer perception of "more pixels = better camera", it is my belief that had Sigma released an SD30 that returned 10 megapixel non-interpolated final full color images, it may have made a big dent in the digital camera market and may have turned the tide provided the technology worked as advertised. As is, it may end up being nothing more than another curiosity. Personally, I wish Sigma/Foveon had made a big leap forward like an SD30, but I also have to realize that true technical marvels take time and often come in small steps. Either way, for me, the SD14 will be an interesting camera that I hope, if nothing else, will help move us forward in the arena of full color capture!

-- Mike Chaney