1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Foveon sensors

Discussion in 'Digital Image Editing & Printing' started by tazio35, Feb 19, 2008.

  1. tazio35

    tazio35 Well-Known Member

    The Foveon sensor: the Betamax of sensor technology. Technically a better system, but doomed to commercial failure. Discuss.
     
  2. beejaybee

    beejaybee Marvin

    Dubious. The fact that the filtering is done "vertically" rather than in a horizontal matrix does not improve the quantum efficiency. In terms of colour rendition and maximum quantum efficiency, the superior system is that used by astronomers: monochrome sensor, long luminance exposure (to obtain maximum detail and low signal to noise), shorter seperate exposures filtered red, green and blue, possibly binned 2x2, to provide colour information. Not really practical for photographing objects with a significant apparent rate of movement, though.
     
  3. tazio35

    tazio35 Well-Known Member

    Doesn't it record all three colours at each point, though? That must be better? Of course, claiming it's a 14 megapixel sensor is bending things a bit...

    I see your point re the astromical method, but what could be done in a single exposure? Could the camera chop up each exposure so that it recorded the colours at all pixels in turn? Foveon's doing that anyway, isn't it?
     
  4. beejaybee

    beejaybee Marvin

    To record colour you need to seperate out R B & G signals (or equivalently C Y & M). So you either seperate by time (changing the filters), by position in the plane (Bayer matrix) or by depth (Foveon).

    The top layer in the Foveon registers R+B+G. Then there is a filter (say cyan) and another sensor (registering B+G). Then another filter (say blue) and another sensor (registering B).

    So we can calculate the red signal by the difference in signal between the first two sensors, the green signal by the difference between the last two sensors and the blue signal from the last sensor alone.

    The blue signal has been filtered twice and also passed through two sensors; many of the blue photons will inevitably be lost and so the sensitivity must suffer ... or noise will be increased compared with what it would be in an unfiltered or single filtered sensor the same size. Meanwhile the red signal will be too great and will have to be scaled down to get a satisfactory colour balance, again wasting a large fraction of the photons.

    I don't know the sequence of the filters in the Foveon sensor, it may be different but the logic will be similar.

    Interpolating colour signals in a Bayer matrix is not a problem provided the sensor coverage is sufficiently dense. Given the relative size of the pixel sensors compared with the resolution of the lenses being used, there really isn't a problem ... anti-aliasing and unsharp masking needs to be applied in any case.

    The theory is interesting, but the Foveon sensor seems to need more development before it can compete with Bayer matrix sensors on sensitivity terms. Meanwhile the inflated pixel counts (which do not correspond to any measure of actual resolution) tend to discredit the technology. It really is a pity that some of the Sigma marketing literature seems to be in total contradiction with the laws of physics; we might treat their claims with more respect if they were not overblown to the extent that they frequently are.
     
  5. FujiSigmaNolta

    FujiSigmaNolta Well-Known Member

    The results speak for themselves. Regardless of its negative points and of what some want to believe, the Foveon does capture more fine detail in its optimal sensitivity settings. Look at examples taken with rather average lenses and see for yourself, how good at capturing detail and colour this sensor is.

    A particularly good example that used to circle around was a photo of a still life including a guitar. The strings of the guitar were always acurately represented in the output image by the Foveon while the Bayer would at times either pixelate it a bit or blended it into other pixels.

    The problem with the Foveon I still believe is the processor behind the chip. Given the constraints of using different colour layers of silicon (and that silicon absorbs light at varying lengths)in the chip I believe the Foveon needs perhaps way much more computing power than what it currently has or may be even possible for the camera (or any camera) at the moment.

    They may be overblowing their horn claiming their chip is a 14MP sensor, but with my SD10s I have been able to open my files as 8MP quite comparably to other cameras and at 11MP not brilliantly but acceptably.

    I am still one to also believe that their crop factor is also one of their major noise hindrances. I believe that if their sensor was perhaps a 1.3x to 1.5x instead of its 1.7x it would have better noise performance. Having said that, the SD14 is much better than the SD10 at high ISOs. There are also tricks (rather debatable, but they work) to have low noise images at high ISO with the Foveons.

    It is a very new technology and it shouldn't be bashed before it reaches its full potential.
     
  6. tazio35

    tazio35 Well-Known Member

    That's sort of what I was getting at with my Betamax comparison: modern multi-head VHS recorders will produce a better quality image than old Betamax recorders, but what if all the years of research had been poured into Betamax? Do Sigma have the same problem with the Foveon sensor in that they are having to do all the research themselves?
     
  7. LargeFormat

    LargeFormat Well-Known Member

    It has always seemed to me that the Foveon is a 4.6MP sensor, albeit that it might punch above its weight compared with a Bayer sensor. It has 4.6 million locations to record information and records three different types of information, RG&B, at each. Point for point the colour information is miles better than the extrapolated information from a Bayer but I can't see how the detail information can be significantly different.
     
  8. beejaybee

    beejaybee Marvin

    ... working at the same native resolution - yes, because the intermediate colour pixels in a Bayer matrix sensor have to be interpolated.

    But is this really an issue? Maybe, if you're dealing with images with big sharp saturated colour blocks in close proximity to each other. For those of us who deal with the rather more subdued colours of natural objects, the Bayer matrix seems to be plenty good enough.
     
  9. LargeFormat

    LargeFormat Well-Known Member

    I don't find any problem with colours from a Bayer but I was anxious to give the Foveon fans the benefit of the doubt. :D
     
  10. Roy5051

    Roy5051 Well-Known Member

    Anything can seem good enough until something better comes along!
     
  11. Mojo_66

    Mojo_66 Well-Known Member

    I got an SD14 a few weeks back, I like the camera, its simplicity, and the images it produces, though I agree it needs lots of light to avoid noise in the shadows, but it has quite a wide dynamic range so it's quite possible to overexpose by a couple of stops and still get a good shot, in some cases it's essential. The Sigma SPP software is very good at getting the best out of the files. As far as sensitivity goes, you'd probably need a tripod sooner than you would using a recent Bayer sensor, but that isn't nescessarily a bad thing!
     
  12. That's not the case.
    THAT'S NOT THE WAY IT WORKS.
    http://www.foveon.com/article.php?a=67
     
  13. beejaybee

    beejaybee Marvin

    Apart from the fact that the filtering relies on the transmission of the silicon rather than being explicit filter layers (which makes no difference to my argument) and I got the sequence wrong (I said I might have) I don't see what you're quibbling at. The physics is invariant.
     
  14. You have the noise in shadows and darker-colored areas, at higher ISOs (in particular 800 ISO and above, but also 400), if you use JPEG conversion inside the camera.
    That's a camera flaw.
    You better shoot in "RAW".
    Then convert with the supplied computer software.
    There are huge differences. As I can read in a review.
     
  15. Bayer sensors (traditional ones) have to declare a different resolution, in honesty.
    Since 2/3 of what they show is "interpolated".
    Simply, doesn't exist.
    It is calculated with the maximum probability, but it is not real.
    You can interpolate a FOVEON sensor too, if you will, and get your "full resolution" image.
    With a Sigma SD14 you can get a full 14 million pixel (resolution) image.
    http://www.dpreview.com/news/0609/06092605sigmasd14.asp
    (4608 x 3072).
    Pixels (photodetectors) in a Bayer are "stacked" with a calculation (two of them, don't exist).
    In a FOVEON, they are truly, stacked.
    As in a film.

    A conventional top 21 mega is a 7 mega, as a matter of fact.
    In fact they say 21 mega pixels (ambiguously), the same as FOVEON does.
    Why should FOVEON declare less?
    They are both incorrect (FOVEON responding, in order not to be penalized), or no one of them is.

    http://www.foveon.com/article.php?a=68 "What's a pixel - Pixel counting definitions".

    Bayer sensors are a disgrace. They destroy reality, then try to rebuild.
    A nonsense process, which is expensive too.
    They are unnatural, with no remedy.
    Take a glance at this comparison: http://www.foveon.com/article.php?a=70.

    Bayers lie, FOVEONs don't.
    Any point of the image, you see the truth.
    The whole image, is true.
    Why go on with that? With such a lie? Artifice?
    Try to think about, "what " YOU ARE LOOKING AT, when you are looking at your photo. A computer calculation. A "maximum probability".
    Leave apart immediate artifices.
    If you like it, keep it.
    I won't do.
    Even if it is a good calculation. "The last image processor, Mk ..".
    What purpose?

    That flower, is interpolated, that face of the girl, is interpolated.. puaghh I don't want that. I'd steer clear of that. Even if interpolation were unnoticeable.
    Even 21 million times. I don't want.
    I have a 21 million points image that was worth living as 7.
    (Or 10 million image that was worth living ..).
    False, with its shiny "superprocessor" telling lies, inventing stories, from its "neighbours". "Having heard that", and placing that as the truth, commercializing that as "resolution".
    No, thanks.
    I DON'T WANT.
    What about you?
    I will kick it, if necessary.
    False.
    Today I can choose, luckily.

    Bayer sensor, you can go to sleep.
    Or to the museum.
    If manufacturers allow to.
    Let FOVEON win (quite difficult. I don't believe they will go knocking on FOVEON's door. Buying a third party sensor. We can only hope).

    Some users reports, on this new technology:
    http://www.dpreview.com/reviews/read_opinions.asp?prodkey=sigma_sd14

    I'm going to buy, a soon as I can.
    SD14 or subsequent.
    What about you?

    Paolo

    I'm looking forward to buying.
    Many Sigma lenses are extremely fine too
    http://www.pixel-peeper.com/lenses/sigma/
    I will continue my inquiry anyway.
    Here another discussion on FOVEONs
    http://forums.dpreview.com/forums/read.asp?forum=1014&message=25036872
    ..
     
  16. El_Sid

    El_Sid Well-Known Member

    Oh God..... another troll!!!
     
  17. TheFatControlleR

    TheFatControlleR :Devil's Advocaat: Forum Admin

    Eeh, calm down pet. You'll give your sen an embolism! :eek:
     
  18. Lounge Lizard

    Lounge Lizard Well-Known Member

    I thought that the proof of the pudding was looking at the output they produced and I can't recall anybody ever screaming that the technology was better than anything else. So much for the theory...
     
  19. Burgy

    Burgy In the Stop Bath

    OK I'll answer the important question. No not in a million years.
     
  20. beejaybee

    beejaybee Marvin

    There's something about Sigma. Perhaps the stuff they put in the silica gel bags turns purchasers into trolls?
     

Share This Page