Guest post: Detailed analysis of the Sigma SD1’s Foveon X3 technology

This is a comment from Dan (@ZDP189) on my last Sigma SD1 post that I think it is worth sharing in a separate thread (you can check also Dan's Fuji X100 review):

Ah, you’ve run into the old Foveon X3 pixel count vs resolution chestnut. Some say Foveon called their technology ‘X3′ because they overstate their resolution by three times.

Foveon’s argument is that Bayer sensors count one sensor element (each of which are monochrome, but tuned to red, green, or blue) as one pixel, so Foveon counts each sensor element as one pixel, even though red, blue and green are stacked on top of each other. In terms of file size alone, this is true.

However, it’s not a count of resolution as we know it. To most people, resolution represents how finely the image grid is laid out. By this measure, Foveon overstates their resolution by three times. The additional colour information is nice because it avoids certain colour artefacts, but it doesn’t represent extra resolution.

The reason it’s less than perfectly clear cut is that Foveon is less susceptible to moiré and the sensor has no anti-aliasing filter. AA filters reduce moiré by slightly averaging out detail, so the sensor is a little sharper without it. Some people claim to be able to see 2/3 of the resolution that Foveon claim. On the other hand, it is a matter of scientific fact that the difference of adding/removing the AA filter does not reduce/increase the resolution by a factor of two, let alone three times, so I think this claim of superior sensor resolution is bunkum. Indeed, many Bayer sensors have no AA filter these days; none of the Super Compacts have one. The X1 does not, the X100 does not. The M8 and M9 do not either. I have never read a complaint about these cameras having moiré issues. These cameras have great resolving power, but I wouldn’t claim that my X100 had 18MP, let alone 36MP.

Furthermore, at this level of pricing, the SD1 system goes against not only the best cameras but the best lenses available. Sigma lenses are notorious for quality issues, back-focussing, corner drop off and lack of corner sharpness compared to top pro lenses. They are not compatible with the other systems; Canon won’t work without serious surgery and while you can adapt a Nikon lens on, you can only do so if you are prepared to give up all electrical functions. If the lens simply isn’t as sharp, it won’t resolve the detail onto the small 23.5×15.7mm sensor at such a high resolution (15.4MP). Simply, the SD1 has 1.7 times the pixel density per area (based on 15.4MP) as the 5D Mark II and 2.9x the pixel density of a D700 or D3S. This will have an impact on optical resolving power, not to mention signal to noise ratio.

Speaking of noise, the Foveon sensors are known to be noisier than Bayer sensors on most entry level cameras. Increase the pixel count without increasing the sensor size and you have the recipe for disaster. Whether done in camera or on a PC, noise reduction squashes detail, potentially a lot more than the AA filter would. You might also argue that as the Foveon has not already had some detail taken out by an AA filter, the detail lost in noise cancelling will be all the more apparent.

In summary, while I would agree that if you count the number of individual sensor elements in the SD-1, you will probably come up with a number of 46MP. However, I expect the actual detail resolving power to be at best commensurate with a camera of 15.3MP, which is nothing special at all.

Look at it this way, if Foveon had simply taken a 15-20MP Sony APS-C sensor and built the camera around that, then would you buy it for even the price of a consumer-grade camera? I probably wouldn’t. The build quality on past bodies has been poor, the control systems clunky, AF poor, processor slow, metering unremarkable and system compatibility is an issue. It’d be trash. You need a super trump card to pull it out of the garbage can.

The only trump card remaining is the colour rendition – not the lovely Kodachrome-esque colours of the JPEGs that really sell the DP line, but the technical limit of what actual RAW files could be processed to. I think a lot of Foveon’s colour reputation is based on what their JPEGs look like or the skill of the user in post processing. I think that most people are not capable of technically identifying the colour difference of the Foveon vs a Bayer, because Bayer interpolation works rather well. I think the only people that can technically justify the SD-1 are people trying to show patterns of colour in very fine detail – possibly scientific and industrial applications. In most cases, including fashion photography and macro photography, the increased sensor resolving power and sensitivity of the high-end Bayer sensors more than outweighs the three layer colour technology of the Foveon X3.

This entry was posted in Sigma and tagged , . Bookmark the permalink. Trackbacks are closed, but you can post a comment.
  • According to dpreview measurent Sigma DP1 & DP2 have same resolution as 10-12MP Bayer sensor. So I assume that DS1 have sam resoltioun as at least 25MP Bayer sensor.

    • Not necessarily; the pixel density has increased a lot; it doesn’t necessarily scale in a lineay way. OTOH, the bigger DSLR lenses might manage to resolve to that density.

      I’d hasten to add that nobody knows what IQ’ll be like until the production models have been thoroughly reviewed and tested.

      The purpose of my comment was mostly about the nature of Foveon’s math versus other people’s math.

      • Tony

        Nothing about your comment, I just wanna say:
        Remember when AMD named their CPU as 3000+, 5000+, 6000+. It was a good marketing strategy at that time since AMD XP 2.16 MHz could run as fast as Intel Pentium 4 3.00 MHz and so on. The same goes for Sigma. It is their marketing strategy and I think it’s pretty fair for the reasons below.

        Think of one square box with 9 small square boxes inside (like Sudoku), and each box works as one Foveon pixel. You can have a yellow X shape inside that box since each box can display thousands of colors. Now, what if each box works as Bayer sensor, can we do the same thing? No, you would need at least 27 boxes. The way that Foveon works is so different which I think it is fair for Sigma to “advertise” it as 46MP.

        Now, about why an “upscale” picture is not looking as sharp as D3x or 1D III. In order to upscale the picture that was taken with Foveon sensor to be as large as Bayer sensor, it has to be done via software. The software has to make a “best guess” of what color should be used for that extra space. Since best guess is still a guess, so the image is lost some detail. But really, ask yourself why you would do that.

        The SD1 has 15MP and each pixel can have its own color! So your print can be very big as it is. Moreover the Bayer sensor also uses best guess (but different method) to crate colors for your print (it’s RGB -> CMYK after all). Think of the square boxes above. You can see that one square box of Bayer sensor could affect 8 boxes around it.

        AA filter? Why anyone even need one when what it does is to manipulate the original picture. It might help if one wants to print an extra large picture from a low resolution image.

        • Tony

          All in all, it is very sad that no one else has the right to produce camera with this sensor. Good IQ but smaller file size, damn.

    • chris

      download the sigma samples.. bring them to 24mp and you will see that the level of detail sucks compared to nikon d3x samples for example..
      no scientific test needed : P

      • I’m not 100% sure the samples were made on a production Sigma body, anyway.

        • person

          They aren’t. Check the EXIF.

          Not a final body and not final firmware. In fact, the sample photos were done with several different versions of the firmware, some quite old at that!

          Bottom line is, we should wait until the finished product is reviewed before jumping to conclusions about the image quality.

          BTW, you are off-base with your assessment of foveon resolution. the SD1 sensor will have just shy of 30MP bayer equivalent resolution. don’t take my word for it, just wait for the tests.

  • Inge-M.

    Dissolvable in a Foveon sensor and proporsjon lens is not difference, so Bayer sensor, but the so is difference is real dissolvable from sensor on ca x2 better so Bayer sensor is, but reason for Foveon use x3 is file size, so the need store is therd time big.

    • Inge-M.

      The big difference is Bayer sensor, have only interpolerstion for make pixels, but Foveon is only direct pixels. The is big difference by dissolvable for Bayer sensor vs. Foveon sensor.

  • Xaver

    Your guest poster has no clue what he’s writing about…

    • Probably not.

      • Parci

        To be honest, this is quite far from a “detailed analysis”… or any analysis actually.

    • Bondi Beach


      IMHO this post should be withdrawn as it appears to be based on opinion and the link from the PR website give the post an air of scientific credibility. I do not have any affiliation with Sigma, nor own any Sigma products, fwiw.

  • ginsbu

    Way to screw up your explanation of the resolution issue! For color subjects Bayer sensors lose resolution in interpolation, so a Bayer sensor will have less resolution than a Foveon sensor of the same size and pixel pitch. Contrary to what you say, for a Foveon sensor “resolution represents how finely the image grid is laid out”, but this doesn’t hold for a Bayer sensor (w/color subjects) since each pixel of the image grid doesn’t collect full color information.

    • That is party true, but the degree to which it is true is far less than the degree to which it is not.

  • Piero

    Indeed guest poster is wrong on the resolution: The resolution advantage of foveon is due to the fact that it does not need to interpolate colors across pixels, the absence of AA filter is only a consequence of that (AA filter helps if you need to interpolate). So for colored details the resolving power is definitely higher than AA-less 15Mp Bayer bodies (A 15Mp bayer body only has 5 Mp blue pixels and 5Mp red pixels).True, SD1 won’t be equivalent to 46 Mp, but it should be around 25 or so.

    The biggest problems of the foveon sensor is the iso noise and more importantly the weird color response in artificial light. The lens problem is also a very important one, and the samples on sigma website indeed show the limits of the lenses.

  • broxibear

    There are some diagrams and further details about the sensor on their site…

    • person

      Everyone needs to cool down and wait for side-by-side comparsions of the final production model by professional reviewers.

      Until then this is all idle banter!

      • broxibear

        Hi person,
        Not sure why posting a link to the foveon site is “idle banter” ?
        Infact why are you replying to my post when I haven’t made any comment at all ?…did you accidentally hit the reply button ?

  • BornOptimist

    What’s the point of this “analyze” – FUD?

  • Sorry but a lot of this is not true. That’s all I’m bothered to say and because why should I care if even the admin doesn’t?

    If anyone’s interested in knowing, they can contact me directly.

  • Pointshooter

    I think this post is a waste of time. No fact and only guesses to the negative side. So better wait until you see and test a physical product before you waste time writing another article.

  • Grev

    As said, lots of misinformation here.

  • Eric

    whoever wrote this should have done a little research first, this is embarrassing. Foveon sensors have proven to have similar resolution to higher megapixel counts, maybe not 3 times but they certainly do. Look at the point and shoots which only have 4.7 megapixels yet can resolve a lot more then other APS sensors at that megapixel range. Common logic would carry that the new SLR will most likely resolve equivalent of beyond 20 megapixels, how far beyond is hard to guess.

    — doesent mean im buying one, but props due where props are due.

  • Joe

    I’d suggest reading the brief article over at Luminous Landscape…

    Michael Reichmann’s been around a very long time, and I’ve found his assessment of various technologies to be generally on the mark.

  • Joe

    Why do you post these useless guess posts? I know this is just a rumor site but let’s show a shred of professionalism here. It read like a typical troll post over at the DPReview forums of which there are many. As usual they are completely ignorant of the facts and likely have never even used a Sigma camera before let alone the SD1. I’m not defending their pricing by a long shot but I’m tired of this camera A sucks because I bought camera B attitude. It’s not based in reality.

    • I did and still own a Sigma DSLR. Have you?

  • TaoTeJared

    I thought it was a good very accurate description of the sensor and the hardships or better stated the inability to compare numbers of Sigma’s Foveon sensor to a completely different technology of Bayer sensors. Sigma’s marketing department certainly takes liberties with describing it’s cameras.

    The difficulty in any numerical comparison would be the same as two farmers trying to compare the numerical results of the size of their farms when one measured it in yards (empirical) and another measured their’s in meters but ignoring what and how they measured it.

    I find it quite humorous when people try to compare the Foveon sensor to a Bayer system as both overstate the actual pixels.

    If you look at it from a reverse stance, that a Foveon sensor is a true 15mp camera since a color is captured at each pixel that makes a Bayer sensor is actually 1/4 it’s manufacture stated value as color is calculated using a 2 pixel wide by 2 pixel high grid. In reality though interpolation (that changes from company to company), a Bayer is this would be about 2/3 the stated resolution.

  • AmirK

    Pretty clueless nonsense by an idiot who has no clue what he is talking about. The 15 MP Foveon sensor would have resolution that is between 20 to 30 MP on bayer sensor
    (closer to 30 than 20), and it’s not just because of AA filter!

    Grow up, moron, and learn something before posting “guest” articles on the internet.

  • The problem with this camera(besides outrageous pricetag) is not the sensor. It’s the fact that it’s doomed because it can only use sigma lenses! Sigma should have made it with a Nikon F-mount or even the Canon EF mount, for God’s sake.

    I wonder if 1 in every 3 Sigma bodies will be soft, like their lenses. That’s gonna be a hot mess, I wouldn’t touch for $800, much less their crazy asking price.

    That is all.

  • Kenji

    I agree he’s lowballing the tested resolution a bit..But still…

    I really and honestly doubt this thing competing well against a D7000 or Pentax K-5 (Both of which have the best APS-C sensors on the market right now) except in some VERY limited circumstances

    While I AGREE that those are pre-production images on the Sigma SD1 site, I also feel that its a massive mistake to post full resolution images if theres still a bug or two to work out of the image processing routines or anything….These are the images that are supposed to be making us go “Wow! I really want this camera” and honestly? I’m just not seeing it…

    I know theres at least one image where I can see clearly visible blotching in the shadows on a woman’s legs and it was only ISO200 or so Thats just unacceptable in this age…ISO200 is -nothing- and shouldnt be….

    On top of it, most images -do- look soft, Whether this is an in-camera sharpening issue or perhaps the lenses the testers using not being up to the job I am unsure, I know there was one where I saw decentering as well, Which doesnt bode well for the lens QC…

    Again, For everyone saying “PRE PRODUCTION” I will restate, These are images that are supposed to entice us to drop $6800 on this camera, These images should make us go “WOW!” with excitement and frankly I do not see $6800 worth of value, In fact in most images I’m struggling to say its as good as my EOS 7D which is a substantially cheaper camera with supposedly far less “resolution”

    Look, I have Sigma lenses, I love Sigma(And I defend the CRAP out of them sometimes), and I was actually excited for the SD1 because I thought it would be a nice competetor against the 7D and D300, But Sigma is instead positioning it against the 1Ds Mark III and D3s/D3x…and honestly, its a fight it cant win, Even with the supposed “Foveon advantage” in color transitions and such a full frame sensor will still have the advantage of lower noise and higher per pixel resolution and sensativity…

    Im not saying the SD1’s image quality will be bad, it wont be, but I just dont see it being better than a good APS-C cam and nowhere NEAR the best FF cams on the market and certainly not on a medium format’s level…

    And lest we not forget the 5DII and D700 are likely to replaced in the coming months at a lower price point, Given how sensor technology has come quite a ways since their introduction I bet they’ll offer even better image quality than their predecessors…I’m betting they’ll close any small gap the Foveon might have in terms of dynamic range and color rendition….

  • tom rose

    Almost everything you can find on the web about digital camera resolution, filters, aliasing etc. is WRONG. There is accurate information to be found, but it takes some searching out.

    The writer of this article does not understand the physics/maths of sampling, or of they do there is no evidence of it here.

    In the first place NO sensor (Bayer or Foveon) can achieve true resolution equal to the pixel density on the chip. The theoretical maximum is half the density. In other words, the best true resolution that a 16 Mp sensor can deliver is 4Mp. The Foveon X3F Merrill achieves this theoretical limit. A Sensor with a Bayer sensor array achieves the same resolution as the Foveon diagonally, but only half its resolution horizontally. On average it achieves 1/SQRT(2) of the theoretical maximum. Strictly speaking it achieves this only for Green and does worse with Blue and Red, but because of the clever demosaicing algorithms Red and Blue performance is better than the raw red and blue resolution.

    To a fairly accurate approximation a 15Mp (45 million photosite) Foveon X3F Merrill chip delivers equivalent true resolution to a sensor with a Bayer matrix and 30Mp, or a little less. The remaining difference is that detail and apparent sharpness with a Bayer sensor varies with colour, whereas the Foveon delivers identical resolution and sharpness whatever the colour, and is less liable to deal badly with some colour combinations.

  • tom rose

    I agree that someone at Sigma must have been off their trolley when they set the SD1’s original price, but in the SD1 Merrill it has come down to levels comparable to any other camera with a similar feature set and build quality.

    It is no longer true that the available lenses are generally inferior to those from Nikon, Canon, or even Leica and Zeiss. Sigma’s latest lenses are very good indeed and the latest ART lenses are a match for the very best.

    There is a lot of useful commentary in this article, even if it seems biassed, but in one further respect it is quite wrong. Resolution is NOT the number of pixels on the sensor, no matter what “most people” might think. That measure is properly and accurately referred to as the pixel count or Megapixel count.

    Resolution is properly measured in line pairs per cm or line pair per picture height. The sensor sets an upper limit to the achievable resolution. Sampling theory proves that in real life resolution down to the individual pixel level is possible only in the very special case where line pairs of 2-pixels width line up exactly with the sensor elements. Obviously that is rarely (never?) the case in practice.

    It actually takes several pixels (at least three) to reliably resolve a line pair on a Monochrome or Foveon sensor. A Bayer sensor needs more pixels because of the Colour Filter Array.

    The actual resolution achieved in an image depends on many more factors than the sensor technology and its pixel count. These include lens aberrations, diffraction, the nature of the subject, the colours in the image and their distribution, what degree of contrast you consider acceptable to distinguish light from dark, the length of the exposure, metering, the absolute aperture used, vibrations induced by mechanical shutters and mirrors, the software that converts luminance values into an image file (demosaicing for Bayer, figuring out true colour values at each layer for Foveon, sharpening algorithms, noise reduction algorithms) and so on.

    Often forgotten is technique. Two things matter. Precise focussing and absolute stillness of the camera during exposure are the two most important. Get those right and an image made with an “ancient” 6Mp Bayer-matrix DSLR will have more resolution than images made with sloppy technique by cameras with MP counts of 16, 24, 38 or 50+ megapixels.

    As for Foveon vs. Bayer “resolution”, the arithmetic is trickier than first seems, and in nay case will not hold for all images. You have to allow for the different resolutions of the three primary colours and for the fact that the theoretical maximum will differ in diagonal directions than in vertical and horizontal. There is also the complicating factor of the strength of any anti-aliasing filter, which spreads the light to reduce the occurrence of artefacts such as Moire.

    Perhaps, in conditions perfectly suiting the Foveon sensor it really can capture the same amount of detail as a Bayer chip with twice as many pixels, i.e. an approximately 40% increase in linear resolution.

    In practice I just don’t see it. I have done side-by-side comparisons of images from the Foveon equipped Sigma SD1 Merill, which returns 15Mp files and a Bayer-equipped Canon EOS 1 Ds Mark ii which returns 16.7Mp files.

    I used equivalent focal lengths (24mm on the SD1, 35mm on the EOS) and equivalent apertures, or as close as possible, for the same DOF (f/5.6 and f/8), and both RAW files subjected to as near identical post-processing as possible (X3F converted to TIFF by SPP then further processed in Iridient. CR2 processed entirely in Iridient).

    I looked only at the central areas of the images so as not to be misled by differences in lens quality. The Canon handles highlights better but I see little difference in the resolution of detail … it is too close to call.

    However, the two systems give a very different colour balance and a different, subjective, “look”. Depending on the subject, sometimes one sometimes the other is more true to life and/or more pleasing to the eye. That is why I have continued to use both systems side by side, even though the Canon is a better picture taking machine in terms of speed of operation, metering, AF speed and accuracy, available lenses, battery life, and toughness. [In the Sigma’s favour are much lighter weight and a smoother and quieter shutter]

    My advice would be to forget the underlying theory, look at the photographic results, at the size and in the medium that you publish, and choose accordingly.

    • tom rose

      I wanted to get to the bottom of this, and after using both the SD1 and EOS cameras side by side for a couple of years, and from extensive testing by my brother and I, we have established, to my satisfaction that, on average, an SD1 image will print well one paper size larger than a similarly sized (in pixel count) Bayer image.

      For example: if a Bayer file of 16.7Mp prints well at A3 then a 15Mp Foveon image of the same scene, prints with almost the same quality at the next size up, i.e. A2.

      We used detailed landscape images for the comparisons. When the scene includes less fine detail the the Foveon advantage is less.

      We reckoned printing to be a more useful test, as it eliminates artefacts caused by the relatively coarse granularity of a computer screen, eliminates irrelevant differences that are visible only when pixel peeping at 100%, and considers what, for me, is the ultimate purpose of the photography … making large prints.

      This equates to a little under 40% more linear resolution which is equivalent to a Bayer sensor with almost 2x the Mp count. If pushed I’d rate the images from the 15Mp images from the SD1 Merrill as holding as much genuine detail as those from a Bayer sensor of 24 to 30 Mp, depending on what is in the frame.

      Sigma might overstate things in their claims, but only slightly.

      A lot of people attribute the unique “look” from Foveon sensors to “microcontrast”, but I do not think that is it. There is plenty of microcontrast in raw files from Bayer sensor cameras, but you have to use the “Clarity” slider in LightRoom to bring it out.

      In fact many photographers seem to destroy much of what microcontrast is visible by increasing global contrast in post processing. Then they blame Canon for producing plasticky or cartoon like images, when the fault is their own lack of knowledge and general incompetence in post processing.

      The big difference, to my eyes, is in acuity. The transition from one object to another in Bayer images always looks a bit fuzzy when you look pixel-peep because of the estimates of colour values that goes on in the de-mosaicing process. Downsizing a Bayer image creates an image much closer in quality to what you get with Foveon.

      On the basis of my experiments, to get image from a Bayer sensor that matches the quality from the SD1 I’d need the Canon 5D ii or Canon 5D iv, depending on what exactly was in the frame.

      I really wish people that don’t know what they are talking about (including me!) would stop spreading their misunderstanding and their prejudices on the Internet. It is ridiculous for the average photographer, with their limited individual experience, to think they understand the technicalities of digital image capture better than the designers and engineers at Sigma and Canon. Either do the research and get it right, or shut up. Please!

      In the end I decided to use my old Canons for everything for three practical reasons:

      – the superior resolution of an SD1 is wasted at the largest size I print (A2). A 16.7Mp Bayer image is good enough.

      – the Canons work faster and are more versatile

      – I am not forced to use Sigma’s “SPP” processor on the RAW (X3F) files

      My brother, on the other hand, decided otherwise. He chooses to use Foveon equipped cameras (SD1M, DP1M, DP2M, DP3M) for all his “serious” photography, despite their operational shortcomings.

  • Back to top