All digital cameras use interpolation

Discussion in 'Digital Photography' started by Charles Schuler, Jan 22, 2004.

  1. A definition of interpolation: to estimate a value of a function or series
    between two known values. Most seem to understand why Bayer sensors use
    interpolation (demosaicking). Many seem to believe that Sigma/Foveon needs
    no interpolation. This is stated in Foveon literature and is often quoted
    on the web. It's not correct. A better statement would be Sigma/Foveon
    needs no demosaicking.

    Stacked photodetectors rely on quantum pumping of a crystal lattice by
    photons plus the fact that shorter wavelengths tend to knock electrons loose
    deeper into the crystal than do longer wavelengths. It's a great idea and
    it works. However, quantum mechanics is about probabilities and it is
    necessary to apply "interpolation" to the Foveon RGB detector outputs to
    determine true color and luminance. Does every "blue photon" penetrate the
    crystal exactly the same and always excite only the blue photodetector? Are
    all Foveon sensors exactly alike? Do they work exactly the same over the
    temperature range to which they are subjected? The answers are obviously
    NO. "Best guess" is what you get.

    I am not knocking Foveon or Sigma. I think the Foveon is a more elegant
    solution than the Bayer mosaic. I believe this technology is going to gain
    momentum because it shows a lot of promise. I don't own/use this technology
    but am certainly watching it and trying to keep an open mind. It's not easy
    thanks to some of the outrageous claims and statements floating around.

    Thanks.
     
    Charles Schuler, Jan 22, 2004
    #1
    1. Advertisements

  2. Charles Schuler

    AArDvarK Guest

    http://www.m-w.com/

    Main Entry: in·ter·po·late
    Pronunciation: in-'t&r-p&-"lAt
    Function: verb
    Inflected Form(s): -lat·ed; -lat·ing
    Etymology: Latin interpolatus, past participle
    of interpolare to refurbish, alter, interpolate,
    from inter- + -polare (from polire to polish)
    Date: 1612
    transitive senses
    1 a : to alter or corrupt (as a text) by inserting
    new or foreign matter b : to insert (words) into
    a text or into a conversation
    2 : to insert between other things or parts :
    INTERCALATE
    3 : to estimate values of (a function) between
    two known values
    intransitive senses : to make insertions
    (as of estimated values)
    synonym see INTRODUCE
    - in·ter·po·la·tion /-"t&r-p&-'lA-sh&n/ noun
    - in·ter·po·la·tive /-'t&r-p&-"lA-tiv/ adjective
    - in·ter·po·la·tor /-"lA-t&r/ noun
     
    AArDvarK, Jan 22, 2004
    #2
    1. Advertisements

  3. Charles Schuler

    noevof Guest

    Foveon is the best and people need to stop with the misleading information.
     
    noevof, Jan 22, 2004
    #3
  4. Charles Schuler

    AArDvarK Guest

    nooooooo ... they are just cheaper in Russia. The
    chip is currently only a really good idea, it is not
    mature yet. And they messed up the original chip
    by putting the micro lenses on it, X3 10M. Now
    it is much more out of color balance.

    Alex
     
    AArDvarK, Jan 22, 2004
    #4
  5. Charles Schuler

    Mike Guest

    Bravo Charles,
    As far as I,m concerned you've got it in a nut shell. I agree entirely to
    your comments in your last paragraph.
    MikeS
     
    Mike, Jan 22, 2004
    #5
  6. Charles Schuler

    Mark Herring Guest

    That simplistic ( and wrong) one-liner earns you a slot in my kill
    file. Looks like a George--talks like a George. By George, it might
    be a. ....................................................MORON
    **************************
    Mark Herring, Pasadena, Calif.
    Private e-mail: Just say no to "No".
     
    Mark Herring, Jan 22, 2004
    #6
  7. Charles Schuler

    YoYo Guest

    Best for the first time photographer maybe... but worst for accuracy
     
    YoYo, Jan 22, 2004
    #7
  8. Charles Schuler

    Dave Guest

    Need a semiconductor physicist to judge its future potential.
    Sometimes what seems like an elegant solution is actually fatally
    flawed by manufacturing issues. In this case I wonder how you inspect
    the green and blue levels of the die for flaws ?
     
    Dave, Jan 22, 2004
    #8
  9. Charles Schuler

    Mark Herring Guest

    Yes---plus the fact that you cannot tailor the spectral response
    curves for each "photosite". I too, think the thing is fatally
    flawed. Perhaps more significant---I have argued that it simply is
    not needed. Bayer works
    **************************
    Mark Herring, Pasadena, Calif.
    Private e-mail: Just say no to "No".
     
    Mark Herring, Jan 22, 2004
    #9
  10. Charles Schuler

    Alfred Molon Guest

    Well no - with Bayer you need to estimate the missing colour values and
    here "estimate" is correct, because you only have one measurement but
    you need to generate three output values - lot of guessing involved.
    With Foveon you can calculate the R, G and B values - no guessing here,
    as you have three input values to calculate the three output values.
    It's more accurate to say that Foveon needs no spatial interpolation,
    while Bayer does - and spatial interpolation reduces the resolution.
     
    Alfred Molon, Jan 22, 2004
    #10
  11. You were doing fine, estimating based upon partially known data, but there
    is even *much* more data than 1 measurement available. Actually 9 or more
    partial spectrum samples can be included in the reconstruction of the Bayer
    CFA RGB output pixel. And even the single pixel is partially known, so there
    is absolutely no need to guess the entire pixel's color and luminance.
    The reconstruction of Luminance is almost perfect while color may, in
    extreme high spatial frequency situations, have a larger error. However,
    since ALL discrete sampling systems NEED a low-pass filter for aliasing
    suppression, that potential color inaccuracy is further reduced.
    No. There are ZERO interpixel clues/measurements available unlike the Bayer
    CFA, so any interpolation between the sampled photosites is a guess.
    to reconstruct a pixel,
    However, the number of pixels in most Bayer CFAs is much larger, thus
    reducing the need to interpolate between RGB pixels to increase the file
    size for printing.

    This year we are faced with many 8+ MP CFA sensors. So far, the Foveon offer
    is stuck at 3.4 MP, meaning that even with an AA-filter the CFA sensors will
    provide superior resolution AND color accuracy.

    Bart
     
    Bart van der Wolf, Jan 23, 2004
    #11
  12. Charles Schuler

    Alfred Molon Guest

    Bart,

    you are simply trying to muddy the waters. Consider the following (CCD
    with N pixels):

    Bayer: N input values, 3N output values - 2N output values are simply
    guessed

    Foveon: 3N input values, 3N output values - no guessing here
     
    Alfred Molon, Jan 23, 2004
    #12
  13. No, they're not. They are estimated, based on a knowledge of what the
    light intensity is at the sites where measurements are available, and an
    assumption that the rate of change of the image is limited (by the
    optics and the AA filter)
    But there is. The 3 colour response curves of the 3 layers are nothing
    like the proper human colour matching functions. So a lot of processing
    and assumptions go into getting any sort of realistic-looking colour
    from the sensor.

    Both Bayer sensors and 3CCD prism blocks have more control over the
    filter spectral response.

    Dave
     
    Dave Martindale, Jan 23, 2004
    #13
  14. Charles Schuler

    Mike Guest

    Does Foveon senser have a AA filter?
    MikeS
     
    Mike, Jan 23, 2004
    #14
  15. SNIP
    Unfortunately not. However, with such an artifact suppression filter the
    camera would produce results that look almost exactly like a 3.4MP Bayer CFA
    image. Since that's not really what people expect for their money, they
    deliberately left it out (which also saves money on Sigma's part).

    Bart
     
    Bart van der Wolf, Jan 23, 2004
    #15
  16. Charles Schuler

    Mark Herring Guest

    See embedded comments

    The word "guess" has always bugged me in this debate. Interpolation
    relies on real information to determine the most likely value to
    insert in the unknown place. To be sure, there is a residual
    uncertainty, but it is much better than a guess
    YESSS!!! The fatal flaw of Foveon, in my view. How can you EVER
    control the spectral band shapes?
    **************************
    Mark Herring, Pasadena, Calif.
    Private e-mail: Just say no to "No".
     
    Mark Herring, Jan 23, 2004
    #16
  17. Charles Schuler

    Mike Guest

    interpolation program to guess what the truth with, regard to the
    Bayer/Foveon wars. Its becoming a religious thing. Time will tell as to
    where the foveon chip is going, if it does what it says on the packet then
    its ok likewise with the Bayer. It all comes down in the end to this. Can
    you tell from which camera a image came from on a double blind test? I think
    you will be hard pressed to know which came from where. Chicken and egg
    situation!
    But I dont want to spoil anyones fun in speculation or haveing a bit of
    displacement activity to fill the time. What ever makes you happy do it as
    long as its not harming anyone, go for it. Dont forget we are a photograph
    site before anything else its the artistic element of the image that counts
    not pixels.
    MikeS
     
    Mike, Jan 23, 2004
    #17
  18. The filter is not normally part of the sensor, it's part of the camera.

    The two Sigma cameras, which are the only cameras that use the Foveon
    sensor to date, do not have AA filters.

    Dave
     
    Dave Martindale, Jan 23, 2004
    #18
  19. Charles Schuler

    Alfred Molon Guest

    Call it "estimate" if you want, but it is still a guess, based on a
    number of assumptions. Besides, if you assume that the image does not
    change between adiacent pixels, you are indirectly saying that the image
    effectively has only 1/4 of the resolution.
    It's a system of three equations with three variables, which can be
    solved deterministically without the need of assumptions.
     
    Alfred Molon, Jan 23, 2004
    #19
  20. Charles Schuler

    Alfred Molon Guest

    Not sure if I understand you. Are you saying that in a Foveon CCD the
    spectral response curves change over time or with the temperature ?
     
    Alfred Molon, Jan 23, 2004
    #20
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.