Sigma vs. Nerd (Part 1: Sensor School)

Anyone who has followed the ULTRAsomething journey for the past several years knows how passionately I’ve lusted for a monochromatic digital camera. A simple perusal of this site should effectively confirm just how little use I have for the color features inherent in modern cameras. Were you to gain access to my private stock of color photos, you’d also see how little talent I have for shooting color. I am a black & white photographer — through and through.

Since the dawn of digital, I’ve known enough about the technology’s inner workings to know that removing a camera’s color capabilities would have significant advantages for BW imaging.  Naturally, I assumed that manufacturers would hear a deafening din of demand from other, similarly-minded BW photographers, and that such a camera would soon arrive on the market. But it never did. Even more surprising was my realization that my BW brethren seemed to have no interest in the idea of a BW digital camera and, worse, some of them even mocked those of us who held fast to such crazy desires. Undaunted, and determined to do something about it, I decided it was time to start lobbying for the camera of my dreams. Since the bulk of my photographic activites demand that I use old-school rangefinder cameras, I started actively pestering Leica to release a BW camera. This time, I apparently wasn’t alone. Because much to my (and everyone else’s) surprise, Leica did exactly that in 2012. That camera — the Leica M Monochrom — is, without a doubt, the most perfect camera ever created for my particular needs and proclivities, with just one single exception: its price.

When Leica first released the Monochrom, they graciously allowed me to borrow one for a couple of months. This experience lead to my three part “Fetishist’s Guide to the Monochrom” review. Sadly, it also increased my already pathological desire to one day own a monochromatic camera, which caused me some serious postpartum depression once I shipped it back to Germany.

After my Monochrom experiences, I had little interest in returning to the quaint and ridiculous process of converting compromised color images into BW. Because of this, I spent the past two years shooting mostly BW film, while waiting patiently either to win the lottery, or for some company to follow Leica’s example and release a monochromatic camera — one that I could actually afford.

So when Sigma announced their new Quattro series cameras a couple months ago, I was intrigued. Sigma uses a unique type of sensor in its cameras — a Foveon sensor. Even though Foveons are designed for color photography (and produce some of the most beautiful color images I’ve yet seen emerge from a digital camera), it’s their untapped potential as monochromatic cameras that interests me.

The more I thought about the redesigned sensor in these new Quattro cameras, the more convinced I became that Sigma was just a couple of marketing decisions away from engineering a monochromatic RAW file format — thus creating a camera with many of the same imaging advantages currently enjoyed by the Leica M Monochrom.

I contacted Sigma and expressed my wishes that the upcoming Quattro camera support a second, user-selectable RAW format — one that basically used only the data collected by the camera’s top sensor layer. Unlike Leica (who had to design a special sensor to remove the ubiquitous Bayer filter), Sigma is already manufacturing cameras with Bayer-less sensors — actually they’re manufacturing cameras with three Bayer-less sensors stacked together. Theoretically, it seems, Sigma could make a true monochromatic camera by simply coding an additional RAW data format (one that retains the luminosity data from the top, high-resolution, low-noise layer), and ignores data from the other two layers (which would normally be used to help define color). Not only would this enable Sigma to ship Quattros that worked as both monochromatic and color cameras, but the simplicity of the monochromatic RAW file should make the data files accessible to industry-standard RAW processors, like Adobe Camera Raw, Lightroom, Aperture, and so on.

Sensor School

Before you can understand why Sigma’s Foveon-based cameras have such potential as monochromatic cameras, you need to understand at least a little something about how the Foveon sensor works — particularly in comparison to the way nearly every other digital camera functions.

It might surprise you to know that digital camera sensors are, in fact, inherently monochromatic devices — capable of measuring luminosity and only luminosity. A camera’s sensor is made up of millions of little photosites — each of which has the singular job of reading how much light is falling upon it. 16 million of these arranged in a grid and spread across your sensor gives you 16 million little points of light — each of which defines a pixel in your final image.

At least that’s how it would work if your camera shot only in black & white. Alas, most people want color cameras. But the photosites used on a digital sensor are nothing more than simple light meters: they’re able to measure the brightness of light, but unable to discern the color of that light. So, in order to create a color image from what’s essentially a monochromatic technology, engineers have employed some rather complex work-arounds.

The most common technique involves placing either a red, green or blue colored filter on top of each and every one of the millions of photosites. Each photosite is then limited to measuring only the luminosity value of the color of light corresponding to its filter. An interpolation algorithm then analyzes all these red, green and blue data points and constructs a color image from the millions of various filtered luminosity values.

The problem with this method is that no single photosite records an accurate representation of the color that strikes it. No single photosite is going to say “I saw yellow light.” Rather, it’s going to say “I saw this much red,” while the site next to it might say ‘I saw this much green,” while another neighboring photosite will say “I saw this much blue.” To deal with this issue, engineers arrange these colored filters in a specific mosaic pattern, from which they construct elaborate mathematical demosiacing algorithms that “guess” the true color of each pixel by examining each photosite in context with all the photosites that surround it. The most commonly used pattern is called the Bayer pattern, though alternate patterns exist (such as Fuji’s X-Trans pattern).

The method works surprisingly well, but it’s not without issues. Specifically:

• The array of colored filters placed above the light-gathering photosites decreases the intensity of light hitting the sensor, thus increasing the base noise level in the image.
• Because no single photosite sees an accurate representation of the amount of light in front of it, image resolution suffers — since it’s the demosaicing algorithm that defines the overall luminosity of each pixel, rather than a pure reading from the photosite.
• Because of the pattern-based nature of image creation, this sort of sensor is prone to inducing moiré patterns in your photos. To combat this, most digital cameras add a second, anti-alisasing filter in front of the sensor — a filter that robs the sensor of even more light and, by design, smears the image details. Some cameras have chosen to forgo the anti-alisaing filter, but the trade-off is an increased probability of moiré patterns appearing in your photos.

Obviously, for a BW photographer, the whole notion of Bayer filters (and their ilk) is ridiculous. The sensor has the ability to record an unadulterated monochromatic version of the scene before it — in gloriously high fidelity, and without interpolation or filtration. But because most people want to photograph in color, we have to endure a tremendous amount of image degradation to support the creation of color photos — an output medium we don’t even desire.

Sigma’s Foveon sensor takes a completely different approach. Rather than a single layer of photosites, the Foveon has three layers, which take advantage of the fact that different colors of light possess different wavelengths. Photosites on the top layer in a Foveon sensor can see every color of visible light. The second layer sees only the green and red parts of the spectrum, since the thickness of the top layer serves to filter out the short-wavelength blue light. The third sensor layer sees only the red part of the spectrum, since the thickness of the top two layers is such that it filters out the mid-wavelength green light, allowing only the long-wavelength red light to reach the bottom. (Addendum: Sigma has pointed out that this is, perhaps, a bit too much of an oversimplification. In reality, all three layers “see” every color in the spectrum, but in different amounts. So, for example, the bottom layer doesn’t see only the red part of the spectrum, it sees mostly the red part of the spectrum. My description is, perhaps, a bit more… umm… “black & white” than what’s actually occurring.)

A sophisticated mathematical algorithm then examines each stack of three photosites and — by analyzing the relative proportion of luminosity reported by each layer — determines the actual color of each pixel.

The benefits of the Foveon method include:

• Increased color purity — The camera is able to determine the color of light on every stacked photosite, rather than approximating each color based on the relative luminosities of several neighboring photosites.
• Increased resolution — Each pixel in the final image contains accurate luminance information, as measured by the sensor’s photosites (in contrast to a color filter array, which must interpolate a luminance value for each pixel).
• Less noise (at low ISO settings) — Because each photosite doesn’t have a colored filter in front of it (nor, possibly, an antialising filter to alleviate the moiré patterns inherent in the demosaicing process), the top sensor layer requires far less signal amplification than a Bayer-type sensor, meaning less noise.

The downsides of the Foveon method include:

• Larger data files — Because Foveon sensors have three times as much data per final pixel, their file sizes are quite large, resulting in increased computational and storage demands. Note that this metric will change somewhat when the new Sigma Quattro arrives.
• More noise (at high ISOs) — Sadly the benefit of having so much unfiltered luminosity data available on the top sensor layer is quickly offset as one increases the ISO speed. That’s because, in order for light to penetrate all the way to the bottom layer, more and more amplification is needed in order to ascertain luminosity values in that bottom layer — and at some point (which appears to be around ISO 400-800 on the current Sigma DP3), the amount of amplification needed to read the bottom layers becomes greater than the amplification needed by a sensor that uses a standard color filter array.
• Limited computer processing options — Because nearly every digital camera uses some sort of color filter array (rather than the Foveon technique), very few computer-based RAW processors can actually read and interpret the RAW files from Sigma cameras. Photoshop, Lightroom, Aperture, DxO, Capture One — none of these popular industry-standard programs can read Sigma files. This means, if your photographic endeavors revolve around one of these programs (and pretty much everyone’s does), then using a Foveon-based sensor is going to add an extra layer of complexity to your workflow. Essentially, you’ll need to use Sigma’s own Photo Pro software to read camera files and convert them to TIF, or you’ll need to use the only known third-party alternative — Iridident Developer.

At this point, you can probably see why (as a BW photographer), I’ve been so intrigued with the idea of Foveon sensors. That top-layer on a Foveon sensor is an unfiltered, full-spectrum monochromatic light gathering machine — the best thing today’s digital world can offer a BW photographer.

Applied Knowledge

So, basically, inside every Sigma camera is a monochromatic camera just begging to get out. And while there are ways to extract all that monochromatic goodness using various bits of computer-based software, it would be much easier and more convenient if Sigma simply offered a monochromatic RAW format — particularly given some of the technical modifications made to the Foveon sensor in its upcoming Quattro series of cameras.

And this is why I contacted Sigma — to offer my assistance in helping them develop and/or test such a RAW format. I have no idea whether or not Sigma is planning such a thing, nor if they’re even taking my suggestion seriously. What I do know is that they wrote back with an offer to let me mess around with the current-generation DP3 camera for a couple of weeks — thus getting up to speed with the Foveon sensor until such time that the Quattros become available.

Although there is very little overlap between Sigma’s current generation of cameras and my own photographic proclivities,  I do have a rather nerdy (obviously) interest in using a Foveon camera, and in learning just how much monochromatic juice I can extract from the Foveon sensor as it currently exists. So, I accepted their offer to borrow a DP3. Besides, who knows? Maybe shooting a camera with such high-caliber color capabilities will reveal within me some heretofore unknown and latent interest in color photography…

… hmm, on second thought, I think I’ll stick with black & white.

The Sigma DP3 is now in my hands, and I will be shooting with it for the next couple of weeks. Sometime after that — once I’ve analyzed the images and worked through some various BW processes — I’ll publish a pair of follow-up articles: Part 2 will focus on the DP Merrill’s usability and capabilities as a tool for the BW photographer; and Part 3 will discuss the handling, ergonomics and general feature set of the DP3 camera.