Foveon Sensors

Discussion in 'Mirrorless Digital Cameras' started by graham john miles, Sep 23, 2009.

  1. After seeing some sample images from the new DP-2, I revisited the Foveon sensor. One thing that puzzles me, is the primary argument that its true pixel count because of the way they are stacked is techically only one third of that advertised. IE 14.1Mp is only 4.7Mp. I can understand the criticism, but should this argument not apply to all sensors? Since CCD and CMOS sensor also have their pixels arrayed in three colours, should their counts be divided by three?. The only difference I can see is that their arrays are set in one plane whereas Foveon's is vertical. Also the link below (add the www) shows a black and white portrait taken with the DP-2. On my monitor, the tonal range appears exquisite and as close to the organic qualities of film as anything I have seen. Is this something only a Foveon could have captured or do people think they could achieve this with a mainstream sensor? PS I have no affiliations with either Sigma or Foveon and us Nikon DSLR's.
    rytterfalk.com/2009/04/20/initial-dp2-review-with-full-size-shots/
     
  2. Most sensors are monochrome but the Bayer filter over the sensor causes each pixel to be either an R, G, or B. On a mainstream sensor the demosaic process looks at adjacent pixels to get an approximate value for the other 2 color values and then merges that data together to form the values for the combined RGB pixel site.
    If you had a mainstream Bayer sensor and a Foveon sensor both the same physical size and the same number of pixels then the Foveon would be able to resolve colors better because the demosaic process for Bayer sensors blends some of the color data together. Since this direct comparison is currently not possible we are left with more touchy feely comments. I've read that a 4.7MP Foveon sensor is about equal to a 8MP Bayer sensor for actually resolving details. It's not as good as a 14.1MP Bayer sensor for resolving details. Of course there is more to resolution than just XY pixel dimensions. The anti alias filter can have a huge impact and more importantly lens choice.
    You can learn more here:
    http://en.wikipedia.org/wiki/Bayer_filter
    http://en.wikipedia.org/wiki/Demosaic
     
  3. Thanks for information and the links Walt. Very illuminating.
     
  4. Milo, you're correct that the pixel dimensions of the DP-2's Foveon sensor (2640 x 1760) yield a total of 4,646,400 pixels - about 4.4-4.6 Mp (depending on what you divide by).
    However, as Walt points out, each of those Foveon pixels has an RGB tuple from exactly the same location. In order to get an RGB tuple from the same pixel location using a Bayer-filtered sensor, you need to combine each 2x2 region into one pixel value, so the "true" pixel-count drops by a factor of four.
    IMHO, the numbers are all marketing hype. If you understand the technology then you can make an informed decision about the characteristics of each camera for yourself. Hopefully the Wikipedia entry that Walt posted will help; here's a few links about the Foveon:
    http://en.wikipedia.org/wiki/Foveon_X3_sensor
    http://mydp1.wordpress.com/2007/11/13/whats-the-big-deal-with-foveon-sensors/
    http://www.foveon.com/article.php?a=67
     
  5. Well, the companies selling Bayer type cameras are sort of misleading you because the commonly accepted definition of pixel involves the R, G and B. And the Foveon people are sort of misleading you by making a corollary to the Bayer sensor misleadment. (Why yes, my English is often tortured.) But looking at photos shot with the DP-2 by people who know what they're doing, what hits me is that even if there end up being fewer pixels, the thing is pretty darn sharp.
     
  6. I would suggest going to DPReview and looking at the DP2 review. Check out the comparison between the DP2 and a Sony DSLR (can't remember the specific model number). The DSLR has a lot more megapixels, but the DP2 has higher overall resolution. Looking at the samples, it is readily apparent that details that are legible in the DP2 samples are not legible in the DSLR samples, even though the DSLR has far more pixels.
    This wasn't always the case with previous Foveon sensors/cameras, but it looks like they're finally starting to deliver on some of their promises.
     
  7. Here is a direct link to the specific comparison I mentioned:
    http://www.dpreview.com/reviews/sigmadp2/page19.asp
     
  8. The Foveon principle swaps color resolution for spacial resolution. Think about it - a Bayer sensor has three separate sensors in the same space as one Foveon site. Still, this subject brings out the Foveon fanboys with their gasoline bombs. It's safer to 'dis the latest Mac O/S.
     
  9. In order to get an RGB tuple from the same pixel location using a Bayer-filtered sensor, you need to combine each 2x2 region into one pixel value, so the "true" pixel-count drops by a factor of four.
    Depending on how one might take that claim, it's either totally false or seriously misleading. Conventional Bayer-sensor cameras don't "combine" four (R-G-B-G) pixels into one color pixel. Obviously the information that each pixel gives the camera about colors other than the one of the filter is limited, and derivation of full-color values depends in large part on data from nearby pixels. But the true, effective, black and white linear resolution of a typical Bayer-sensor DSLR is about 75-80% of what its pixel count implies. For example, if the sensor has 3000 pixels on the short side, it can probably resolve about 2300 black-and-white lines on that side. If as claimed the camera simply combined the pixels, the effective resolution would be less than 1500 lines, probably significantly less.
    The reasons for less than pixel-count resolution are several, including the anti-alias filter, the RGB de-mosaic, and the Nyquist limitation. In terms of real-w0rld perceived detail, it's usually the black-and-white resolution that matters most (back in the day, analog color television had much higher resolution for the luminance (B&W) signal than the chroma (color) signal). And the Bayer sensor and RGB de-mosaic issues matter less for the detail that typically matters most.
    I do agree that all of the camera makers fudge their stats, but the Sigma/Foveon claim are both more misleading and (IMO) more intentionally so than are the Bayer-sensor camera manufacturers' claims. The Sigma cameras are truly 4.7 MP, and their effective resolutions are (IMO, based on published test reports) generally comparable to most DSLR's of 9 or 10 MP.
     
  10. About a year ago I said Foveon is 2x better in comparisons with Bayer. So 4.7 * 2 = 9.4 agrees with Dave's 9-10 Mp equivalency. Did anybody note that in Dpreview's test of the DP2, the Olympus EP1 is the clear winner for clarity, but the DP2 creates a much more 3D effect? The Foveon sensor really is amazing. And I would rather have a crystal-clear 4.7 Mp than a crappy 10 Mp that needs to be downsampled to look good. Faster editing, consumes less storage. Too bad the DP2 does not meet my needs.

    By the way, I believe 200% zoom is a much better way to compare images than is printing.
     
  11. Dave, thank you for that calm and reasoned post. As someone who's written Bayer demosaic algorithms and designed industrial cameras using both Foveon and Bayer sensors, that whole "you need to combine each 2x2 region into one pixel value, so the 'true' pixel-count drops by a factor of four" spiel had me on the merge of a major rant, combining math and rudeness.
    That aside, any arguments about tonal range have mor to do with how a particular image was exposed than the sensor use to capture it, and the "3D effect" is again more lens and composition than sensor. We didn't have films with a "3D look" back in the film days, we had lenses. The Nikon 105mm f2.5 was always a favorite, and of course, borrowed Leicas.
    Kudos to Sigma (a lens company "dabbling" in cameras) for "leading with the lens" on the DP2. It would be a fine camera, no matter what sensor were behind it.
     
  12. Joseph, glad you like the post, although I will have to admit having enaged in a rant or two myself!
    The funny thing with me personally is that, as much as I have criticized Sigma's Foveon-sensor-camera marketing, which I regard as borderline fraudulent, I actually like the idea of the Foveon sensor (and for that matter, the idea of the DP1 and DP2). It just seems that so few (comparatively) are produced that the Foveon technology cannot keep up with the Bayer sensor technology (regular CMOS or CCD), likely due to the huge disparities in research and development expenditures. (And DPReview's take on the DP2 gives a pretty good summary of how it is way behind many of its competitors.)
    From the samples I've seen (I have never used a Foveon-sensored camera myself), it does appear that, at least for certain types of subjects, Foveon sensors do seem to produce more pleasing results. It's just that, to me, their major limitations outweigh their advantages. If anyone finds a Foveon-equipped camera the best solution for his needs, super, have a Foveon.
     
  13. Dave, Joesph: sorry for the oversimplification in my post. I did put the word "true" in quotations after all.
    Feel free to flame on - you won't hurt my feelings (or likely use math I'm not familiar with...).
     

Share This Page

1111