Why don't film cameras use anti-aliasing?

Discussion in 'Digital Cameras' started by K2, Nov 28, 2003.

  1. K2

    K2 Guest

    To those who keep saying Foveon chips (or at least their incarnation
    in Sigma cameras) are inferior because they LACK anti-aliasing, I ask
    this question:

    Would the output of a 35mm or 4 x 5 film camera be improved if the
    light was blurred (anti-aliased) before it reached the film? Should we
    start using fuzzy lenses to mimic the "superiority" of Bayer sensors?

    The point is that Foveon sensors don't need, want or desire
    anti-aliasing in the first place. The lack thereof is a freedom, not a
    limitation!

    K2
     
    K2, Nov 28, 2003
    #1
    1. Advertisements

  2. You need to look up the word "aliasing" in a book or article about discrete
    sampling or digital signal processing.

    Briefly, aliasing is a phenomenon that occurs when a frequency higher than
    the highest that the system is capable of representing is sampled. Aliasing
    shifts those frequencies to frequencies that the system is capable of
    representing.

    Aliasing only occurs in discrete sampling. Since film isn't a discrete
    sampling system, aliasing doesn't occur.

    Aliasing is bad because you can't tell the difference between real
    information and aliased information. So you can't fix it after the fact.
    No, anti-aliasing is required for correct imaging. If you don't want cat
    hairs that look like beads and barbershop poles, you have to antialias.

    David J. Littleboy
    Tokyo, Japan
     
    David J. Littleboy, Nov 28, 2003
    #2
    1. Advertisements

  3. Nope. Because the sampling rate is so high. I guess your lack of
    knowledge of sampling theory explains why you defend Foveon's low-
    resolution design.
    Sensors don't "want or desire" anything. They are devices, not beings.
    Actually, it's a significant flaw, considering the relatively low sampling
    rate. The weird false resolution of Sigma images demonstrates this.
     
    Albert Nurick, Nov 28, 2003
    #3
  4. K2

    K2 Guest

    I know full well what it is. I was trying to be slightly humorous with
    this topic.
    When "the system" in this case is your computer monitor at a limited
    resolution, why not improve the monitor instead of degrading the
    camera sensor? That's been my big point in this discussion.
    But Foveon sensors are close enough that it doesn't really matter,
    provided your monitor is set to a high enough resolution. We've been
    over this before.
    Once again, that's much more a function of your monitor settings, not
    the sensor. Ideally each pixel on the sensor would be infinitely small
    but you have to be pragmatic. ANY digital image will show some sort of
    jagginess if you blow it up enough, so it's par for the course and not
    worth obsessing over.

    K2
     
    K2, Nov 28, 2003
    #4
  5. K2

    K2 Guest

    The only thing lacking here is your understanding of my dry humor!
    I made that statement because certain individuals in this group keep
    insisting Foveon (or rather the Sigma implementation) "needs" to add
    anti-aliasing.
    They look quite natural to me and only weird or false if you insist on
    comparing them to outdated Bayer fuzziness. Try to appreciate change
    for a change!

    K2
     
    K2, Nov 28, 2003
    #5
  6. K2

    K2 Guest

    I think digital differs from film in that you shouldn't be expected to
    blow an image up beyond normal size and not see some stair-stepping.
    It's the nature of the format. There must be a practical limit to how
    small you can make a sensor pixel and have it gather light accurately.

    Film grain is apparently on the order of 2 microns and Foveon X3
    pixels are said to be 9 microns. I guess that adds some validity to
    your point RE the X3s current implementation. But the pictures look
    quite good to me as is. If they reduce the pixel size further, which
    I'm sure they will, I don't think you'll have much of a case for
    anti-aliasing.

    K2
     
    K2, Nov 28, 2003
    #6
  7. K2

    K2 Guest

    I want to add that the random distribution of film grains would
    confuse anti-aliasing software anyhow. It wouldn't know where one
    grain started and the other left off. But this whole post was mostly
    sarcastic. :)

    K2
     
    K2, Nov 28, 2003
    #7
  8. K2

    Graeme Guest

    I made that statement because certain individuals in this group keep
    It does need to be added.
    <PLONK>
     
    Graeme, Nov 28, 2003
    #8
  9. You say things that indicate that you don't understand what aliasing is. And
    you continue to in the following.
    Your understanding here is incorrect. The "system" in this case is not the
    computer monitor, but an array with a fixed number of data points, i.e. the
    image output by the camera + software.
    Then your point is misplaced, mistaken, and wrong. The monitor has nothing
    to do with it. The aliasing artifacts occur at the point the sensors measure
    the image. They occur because Sigma left out the AA filter. To prevent
    aliasing from occurring, you must filter the image _before sampling_ to
    remove frequencies above what the system is capable of rendering.
    This is incorrect. Aliasing always occurs if the higher frequencies aren't
    removed. If the image is defocused, or the lens is not sharp so that the
    image contains no frequencies at or above the Nyquist frequency, then you
    won't see artifacts, but the image will have the same resolution as it would
    have had there been an AA filter in place and it was focused and the lens
    sharp.

    The idea that a Foveon sensor can resolve more than a Bayer sensor can is
    simply wrong.
    Then you didn't understand it. The SD9 and SD10 create images that are full
    of aliasing artifacts. However they are displayed or printed. The data's
    bad.
    This is wrong: the aliasing artifacts occur in the image, not the monitor.
    No, it won't if (a) you don't try to represent frequencies above 2/3 of the
    Nyquist frequency, and (b) you upsample in a sensible manner.

    David J. Littleboy
    Tokyo, Japan
     
    David J. Littleboy, Nov 28, 2003
    #9
  10. This is completely incorrect. There's no need to see stairstepping with
    digital images, ever. Quite the contrary. If you start with an image that is
    bandlimited to roughly 2/3 of the Nyquist frequency, and upsample in an
    intelligent and careful manner, you get images that are as smooth as smooth
    can be.

    This is one of the main reasons people like digital: this plus the lack of
    noise in dSLR images means that you can blow them up as large as you like.
    Film gets ugly quite quickly at enlargements much over 10x.

    David J. Littleboy
    Tokyo, Japan
     
    David J. Littleboy, Nov 28, 2003
    #10
  11. I agree with you David, but apparently some people like high pixel to pixel
    contrast. Anything else is fuzziness. It is odd to me that they believe
    that all sharp edges in the universe line up with the edges of the pixels in
    their sensors, but they do and they are willing to put up with the
    sharpening that it takes to achieve it.
     
    Gherry Bender, Nov 28, 2003
    #11
  12. Preddy -troll-o-meter

    0 1 2 3 [4] 5 6 7 8 9 10
    < >
     
    Betty Swallocks, Nov 28, 2003
    #12
  13. K2

    Don Stauffer Guest

    Aliasing only occurs when sampling at a constant pitch or inteval. Even
    though one can argue that film is a sampled data system with the grains
    doing the sampling, the pitch of the grains on the film is random.
    Random interval sampling does not create aliasing.

    There have been attempts to make random sampled electronic cameras, but
    this so far has never proved practical on staring array focal planes
    like CCD imagers. It has been implemented with image dissector cameras,
    however.
     
    Don Stauffer, Nov 28, 2003
    #13
  14. Yowza. Another Preddyesque kook.
     
    Albert Nurick, Nov 28, 2003
    #14
  15. K2

    Matt Austern Guest

    I suspect it's the "randomly placed" part that's relevant.
     
    Matt Austern, Nov 28, 2003
    #15
  16. K2

    JPS Guest

    In message <>,
    I ask you this question: are you really this clueless?

    Film does not have a rigid grid of samplepoints. It is impossible for
    visible aliasing to occur on film.

    It *is* possible on a bayer sensor (without anti-aliasing). It *is*
    possible on a Foveon (without anti-aliasing).
    Take a close look at the B&W spokes in SteveGeorge's PNG.

    --
     
    JPS, Nov 28, 2003
    #16
  17. K2

    JPS Guest

    In message <>,
    Backpeddling is fun, isn't it?
    --
     
    JPS, Nov 28, 2003
    #17
  18. K2

    JPS Guest

    In message <>,
    It certainly does. Not as much as a bayer CFA, but it needs it
    nonetheless, and SteveGeorge's B*W wheel PNG amply demonstrates it.
    --
     
    JPS, Nov 28, 2003
    #18
  19. K2

    Rafe B. Guest


    And they're absolutely correct. It's your own
    understanding that's amiss.


    rafe b.
    http://www.terrapinphoto.com
     
    Rafe B., Nov 28, 2003
    #19
  20. ?

    It's a bit more complicated (and expensive) than that, but essentially
    that's it.

    The only current digital cameras that do not have an antialiasing filter are
    the Sigma SD9 and SD10, and the Kodak 14n. Some of the medium format digital
    backs (e.g. the Kodak back) lack an AA filter as well. (At least one of
    those has an optional AA filter that's user attachable/removable.)

    David J. Littleboy
    Tokyo, Japan
     
    David J. Littleboy, Nov 29, 2003
    #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.