Canon 10D's anti-aliasing filter question, and AA filters in general

Discussion in 'Digital Photography' started by William Wallace, Jan 15, 2004.

  1. I know about Nyquist's with regard to time varying signals. The
    sampling theorem applies to PAM, and assumes infinite precision zero
    error analog samples at a specified jitter free frequency. The
    sampling theorem is useful PCM, because quantization noise is well
    understood.

    One real-world constraint with anti-aliasing filters is that it is
    difficult to design filters that approach the ideal brick wall
    approach. This is solved a number of ways, e.g., over sampling and
    using an AA filter with a cut-off frequency less than 1/2 the sample
    rate, say 1/4 or 1/8 the sample rate, and a step slope, say 60
    dB/decade.

    But with regard to imaging, I would like to better understand AA
    filters.

    Do they set the cut-off frequency at 1/2 the sampling frequency?
    E.g., at 2 pixels? How is this done? Is it a brick wall? Does it
    operate in two directions (horizontal and vertical) or does it also
    work diagonally? Are they optical or digital?

    And getting to the Canon 10D, I have noticed that it does not capture
    my son's eye lashes too well...sometimes, it just erases them. Is
    this due to a poor AA filter design? Bad luck (e.g., maybe all of the
    eyelashes are pointed straight at the camera)? A feature of the 10D
    sensor? Strangely, a single fine hair laying across my wife’s
    face is captured? Is the 10D using an algorithm to not filter hair of
    a certain length? (A neat trick). Maybe my wife's hair is just
    thicker than my son's eye lashes.
     
    William Wallace, Jan 15, 2004
    #1
    1. Advertisements

  2. William Wallace

    andrew29 Guest

    Optical. It's usaually done by stacking multiple layers of a
    birefringent crystal such as Lithium Niobate.

    A birefringent material splits a light beam into two parallel
    components, depending on polarization. The distance between these
    components is less than or equal to the spacing of the pixels. Stack
    two of these crystals at 90 degree angles and each beam is split into
    four components. This is quite a precise technique, and not just a
    simple blur.
    Well, that depends on how close the size of the hair is to the
    resolution limit of your camera.

    Andrew.
     
    andrew29, Jan 15, 2004
    #2
    1. Advertisements

  3. William Wallace

    KenP Guest

    As a different approach, the 14n uses no physical anti-aliasing
    filter, simply allowing the blur circle of the lens image (such as it
    is) to accomplish the anti-aliasing. This permits higher possible
    resolution, but in the case of exceptionally sharp lenses and distant
    picket fences (for example) the aliasing can become evident in the
    image (if not on the print). Perhaps as image sensors become even
    more dense, the need for anti-aliasing blur filters will disappear
    altogether.
     
    KenP, Jan 15, 2004
    #3
  4. Nope. Just as in audio, such a thing doesn't exist. The Fourier
    duality still applies. So every AA filter is a compromise. Kodak used
    to supply a removable filter for their DCS cameras; this was
    apparently more aggressive than filters built in to other cameras, but
    could be removed for ultimate sharpness, at the risk of aliasing. As
    another poster pointed out, Kodak's most recent camera does without
    any filter, but is susceptible to aliasing, e.g. on tweed jackets and
    the like.
    Optical; as you know, once aliasing has been introduced in sampling,
    no digital processing can remove it.

    Fortunately, however, the aliasing is most objectionable in the color
    components (at least for non-Foveon cameras), and with some insight
    into human perception and the 3-color sampling process, as well as
    perceptual limitations, can often be masked or filtered out by
    software.
    Is it that a bright object against a dark background is more visible
    than the opposite?
     
    Stephen H. Westin, Jan 15, 2004
    #4
  5. William Wallace

    Don Stauffer Guest

    I don't see how that produces higher res than a specially designed sharp
    cutoff filter. Most lens aberrations have a very gradual slope near
    cutoff (slope to mtf), so are not a sharp cutoff. If the normal lens
    aberrations are doing the filtering, I am surprised that this is sharper
    than something designed from scratch to produce a sharper cutoff.

    I must also admit I have never tried lens design with birefringent
    materials, so I have no idea how hard it is to produce a sharp cutoff
    that way.
     
    Don Stauffer, Jan 16, 2004
    #5
  6. William Wallace

    andrew29 Guest

    I'm not sure this qualifies as a different approach. It's a cost
    saving measure that happily ensures better performance on resolution
    tests. Ideally, you want high contrast right up to the Nyquist limit,
    and then a sharp cutoff. If lenses really did all the necessary
    antialiasing they'd lose a lot of contrast below the Nyquist limit,
    and such lenses are rightly criticized as soft. But I'm not a lens
    designer; maybe lenses with the right characteristics can be made.
    Perhaps, although we've a way to go yet. Besides, in many areas of
    photography people are looking forward to the day when high
    performance digital systems can fully replace medium- and large-format
    film.

    Andrew.
     
    andrew29, Jan 20, 2004
    #6
    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.