Sigma has, for years, blazed a daring path with a unique sensor technology. It has many fans and I’m one of them.
A camera sensor is a photon counter. It counts how much light reaches any given pixel location (sometimes called sensel”) on the sensor.
Some cameras record only monochrome information — just pure black and white information, such as the Leica M Monochrome cameras and monochrome converted cameras.
With a red or green or blue filter above the pixel, each pixel only records information about one color. To get full color information, the image processor must go through a process called “demosaicing” — literally taking the mosaic of colors and by using information from neighboring pixels, figure out how much red and green and blue information belongs at each pixel site.
This leads to something called “Bayer Blur.” Each pixel only records partial color data & it needs to take an average of neighboring pixels to get full color data — this averaging leads to some blurring.
Red light has a longer wavelength than blue. Foveon exploits this by reading how deeply a photon penetrates into the sensor to measure full color information at each pixel location. The below image is a simplification, but gives you an idea of how the Foveon sensor works. (Each layer isn’t capturing just a single color & complex math is needed to extract RGB data.)
Since it’s able to capture full color detail at every sensor location, it can be as sharp as a monochrome sensor, but also record full color data.
Example Photo — Bayer
This photo was taken on a Sony A7R2 using the Sony 85mm f/1.8 lens. This is an extreme crop of the image — this was a half body shot from the waist up. Also — congrats to Sony on the excellent Eye autofocus.
Despite being shot wide open, the image is very sharp (at least the parts that are in focus). If you zoom in (see below) you can see a bit of color moire artifacts in the parallel lines of hair — but that’s such a minor detail. Especially when each hair is only maybe 2 pixels wide.
The excellent image quality shouldn’t be a surprise — the A7R2 is state of the art technology. A 42 megapixel sensor without an AA filter, but I was very surprised at the amount of detail as I kept zooming and zooming — in that respect it felt like editing a Sigma file.
It’s worth noting that I spent a lot of time in Adobe Camera Raw getting the colors to where I liked them.
Another sample photo.
The above image was shot at f/5.6, so the lens is working with the sensor for sharpness. This is, again an extreme crop — this was a full body shot (model is sitting on a stool). Some sharpening has been applied.
These are our reference photos for what a top of the line Bayer sensor can do.
Example Photo — Merrill
This photo was taken with a Sigma DP3 Merrill camera. No sharpening (other than Sigma Pro Photo defaults) was applied, in fact no edits at all, this is just what Merrill photos look like. (The way I lit the scene has a lot to do with the way the image appears as well.)
Notice how much more “realistic” this image looks. To me a Merrill file feels like each pixel really exists in 3D space. Skin feels solid in a way that I’d never experienced with a bayer sensor camera.
Eyes and mouths were one of the most interesting things to me going from Bayer to Foveon — the lines are so sharp. I wasn’t used to eyelid simply stopping and eyeball beginning. The 42 megapixel sensor in the Sony A7R2 is the closest I’ve gotten to this feeling, and to get to that required a bit of sharpening, which introduced an artificial feeling to the image.
This photo was taken on an APS-C sensor camera with a 75mm equivalent lens at f/5.6 , and was a head and shoulder shot rather than a half body shot — so of course some of the depth and sharpness will come from that, but I’ve never seen something like this from a Bayer sensor camera.
This is a 15 megapixel sensor. Exported at “double size”, this is a 60 megapixel file that — on the pixel level — feels sharper than the Sony 42 megapixel file. That’s because it has great microcontrast. As distinct from sharpness, microcontrast is the ability for each pixel to retain high contrast from a neighboring pixel. See my discussion on the topic for a more technical explanation.
Sigma Quattro Technology
As much as I — and many Sigma aficionados love the Merrill sensor, it has limitations. It’s not very sensitive to light, especially in the “red” channel. Those photons have a long way to go through that silicon to reach the “red” layer. If there isn’t enough light, the sensor is prone to doing weird things with color. Also because of the complex math involved to process an image, it takes about 8 seconds to write a file. You take a photo and wait 8 seconds before you can see it.
So Foveon devised the Quattro sensor. They combine the lower layer information in a 2x2 grid. Larger area means more photons can be counted, means more accurate color data can be recorded. I believe Sigma and others say this adds about 1 stop of light gathering capacity — meaning more sensitivity to light and more accurate colors, especially in low light/shadow areas.
File write times are now 4 seconds instead of 8.
Foveon’s engineers argue that the eye sees the most detail in the green spectrum, so it was OK to combine the lower layer information. Most of the detail would be retained by the top layer, and you’d get more color accuracy, low light performance and faster file write times.
The eye’s sensitivity to green is also why there are twice as many green pixels in the Bayer color filter array.
Many Sigma/Foveon enthusiasts felt this was a step backwards — combining data from neighboring pixels is what Bayer sensors do. Much debate was had over whether or not Sigma lost “that Foveon look.”
Example Photo — Quattro
So — is Quattro a step backwards for Sigma/Foveon? Yes and no.
The image is still distinctly Foveon, but has lost some of that microcontrast “pop” that the Merrill image has. The Merrill image is easy to look at — it feels real, like it’s really there. The Quattro image feels like there’s just a little bit of extra haze or blur to it (some of this is from exporting at “double size” — which pushes the image quality to the limit).
The Quattro and Merrill photos were taken under basically identical setups. Same lighting, same aperture (within 1/3 stop). The Quattro image “Portrait” color mode in SPP feels overexposed, so I applied -0.5 EV compensation to it. I think the lighting on the Quattro image was a bit softer than for the Merrill (larger light source).
The Quattro sensor in the DP3Q is 20 megapixels. Exported at “double size” it’s a 78 megapixel file — a massive, massive file — pixel peeping on a 78 megapixel file is sort of cheating. It’s like criticizing the chisel work on George Washington’s nose on Mouth Rushmore.
Zooming out to the actual image, it has the Foveon feeling of actually existing in 3D space. The hands are actually there, the face is actually there. The “bayer blur” robs the scene of the sense of reality.
As a portrait, it’s actually very pleasing. A nice compromise between the overly-detailed look of the Merrill and the overly-blurry look of the Bayer.
Here’s another photo — Foveon sensors love denim — lots of juicy details for them to pick up on.
I find that Portrait color mode images in SPP need about a half stop negative exposure compensation — they seem too bright otherwise. The colors are nice & the details are good — but it lacks some of that microcontrast “pop” of the Merrill.
[edit: More specifically it seems like the blacks are lifted a bit — by about a half a stop, and that pushes the shadows and midtowns up. The highlights are in place, but everything else seems too bright.]
Below is the same image in Monochrome mode — where most of the data comes from the top layer (or maybe they just add the total amount of light together from all layers?). Suddenly that blurred feeling is gone — it’s sharp and contrasty — and feels like a Merrill photo of denim.
Below is the original color image, with the monochrome image overlaid on top with the blend mode set to Luminosity — so the contrast is coming from the monochrome image and the color is coming from the color image.
This I like a lot. [edit: actually I love this look, it’s worth the effort.] Not only is the softness gone, but the colors are more accurate as well. (It’s a dark denim jacket.) It’s still not as extreme with the microcontrast as a Merrill image, but it’s much better than the default SPP Quattro output. (and the Merrill images may have overdone it with the microcontrast…)
It’s worth noting that no exposure compensation was applied to the monochrome image. It’s a bit overly contrasty, but that could be worked with — or left alone. By reducing the opacity of the monochrome layer alone there’s a lot of options for how to process this image. Or you can start with a less contrasty monochrome image.
Could this be the secret to making Quattro files look good — more like classic Foveon files? Use the monochrome output from Sigma Pro Photo for luminance data when post processing?
For comparison — the Portrait color mode with no exposure compensation, the Monochrome layer atop and the monochrome image alone. Remember — this is a dark denim jacket, not a light denim jacket.
Perhaps it’s Sigma Pro Photo’s portrait color mode that makes the images feel soft. Since the red pixels are binned into 2x2 — if I export at double size mode, so they’re more like 4x4 — if portrait mode relies heavily on data from the lower levels that would cause blurring (it’s common for camera & software manufacturers to use more red data in “portrait” color modes as skin looks smoothest in the red channel).
If you go back to the diagram of how the Quattro sensor works, “resolution” data was supposed to be taken from the top level. So why would my mimicking that behavior make the photo suddenly look more Foveon like? I suppose only Sigma and Foveon’s engineers can explain exactly what’s going on when Sigma Pro Photo processes color.
Again these exports were using the default settings in Sigma Pro Photo.
Fuji — who has been a pioneer in sensor technology for decades — has rearranged the color filter array into what they call X-Trans.
Notice that while in a 2x2 grid (or 8x8 grid) of Bayer pixels, you have 50% green, 25% red, 25% blue. In a 3x3 grid of X-Trans (the proper way to count) you have 5 green pixels to 2 red and 2 blue so 55.6% green, 22.2% red and 22.2% blue. The actual ratio of colors is different.
Almost all Fuji X series cameras have X-Trans sensors in them (some have Bayer sensors).
What can I say — I’m a convert.
The colors are great right out of the camera & that, to me, is one of the most important things — being able to hand files over to a client with as little time spent in photoshop as possible is the dream.
Is it X-Trans or just Fuji’s color algorithms? I don’t know. The colors are great even when processing raw.
Example Photo — X-Trans
While I love the Fuji X-Trans images in general and it’s become my default studio camera (along with my Sigmas and Sony A7), the X-Trans isn’t perfect.
When it comes to sheer pixel-peeping detail, the 24 megapixel sensor in the Fuji X-Pro2 can’t keep up with either the 42 megapixel sensor in the Sony or the Foveon sensors — but it easily gives the 24 megapixel Sony A7 a run for its money, and for my money, the Fuji is sharper and more micro-contrasty.
Photos out of the X-Pro2 seem “solid” to me in a way similar to the Sigma cameras. Remember the second Sony file (with the red hat) was sharpened — this was not. Perhaps it’s because of the X-Trans sensor has 2x2 sections of green pixels, allowing it to pick up more microcontrast.
The colors are easy to work with & easy to hand over to clients, basically unedited. The less post production I have to do, the better. The Sony A7R2 required a lot of work to get the files to look good.
When the X-Pro2 falls short is the reds — there’s fewer red pixels & they’re spread further apart than on a Bayer. While every green pixel on a Bayer has 2 red pixels from which to derive color information, on the Fuji, each green pixel has only 1 red pixel from which to infer colors. This means skin tones can end up a bit “waxy” and areas with heavy red detail (like the model’s hair) can get lost.
This is mostly a concern for portrait photographers, though.
Also the sharpening algorithms in the X-Pro2 are strikingly similar to the sharpening algorithms of the Sigma Merrill camera (but also very different). I make a direct comparison of them here.
Conclusion & Additional Notes
Different cameras have different personalities, and I use each in different ways that makes sense with their personality.
Here are my “tasting notes” for the Sigma and Fuji cameras mentioned in this article.
- Sigma Merrill — Crunchy details, which can be good or bad depending on your taste. There’s a distinct “3D” quality to the pixels that goes beyond microcontrast.
- Sigma Quattro — A good compromise between the Merrill crunchy and Bayer smooth, though for portraits/portrait color modeI find it takes work to get the files to look good.
- Fuji X-Pro2 — My default studio camera. I only own autofocus lenses for Fuji. The images it renders feel very much like it belongs in the Foveon family, even if it can’t quite compete on resolution.
- Fuji X-Pro1 — Much more Bayer-like in its rendering, but in a good way. It reminds me of my beloved Nikon D7000 (I think the underlying sensor was from the same chip?) It sort of is to the X-Pro2 as the Quattro is to the Merrill — a smoother rendering with less sharpening.
For me the best all-around camera that I own is the Fuji X-Pro2. The colors are great right out of the camera. The 24 megapixels are “good enough” unless you’re pixel peeping or need to crop, and they’ve got a Merrill quality to them.
The love of Fuji colors and desire for more resolution has made me seriously think about Fuji’s medium format cameras.
I’ve sold my Sony A7R2 and lenses. Though I loved the images I produced with it & know I can’t replicate that look with my current equipment*, I felt like I was fighting the cold/harsh Sony colors in post production and it wasn’t worth it to keep for that reason.
And yes I calibrated the camera using an X-Right colorchecker, with both the X-Rite software and the Adobe DNG Profile Editor software. When I spend more time calibrating the camera than anything else while editing photos — it’s time to sell the camera.
Sony A7R2 — Mila Bog
Sigma DP3M — Jenyi Lee hair and makeup by Holly Whitehead
Sigma DP3Q — Brandy Godsil, Alexis Kristiana (denim jacket)
Fuji X-Pro2 — Yaya Luciano hair and makeup by Kindsay Kastuk
Photographer — Me.
My apologize to everyone involved for posting unedited photos.
About the Blog
I’m a photographer in New York City, specializing in portraiture, but I take lots of other kinds of photos as well. I was practically born with a camera in my hand — my first photoshoot was at age 8 with my action figures. While I have no engineering qualifications — and white papers put me to sleep — I do like to geek out over camera technology & share my knowledge. I also teach music theory and literary theory. By day I’m an advertising executive.