Bayer with NO anti-aliasing (Kodak Pro 14n)

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

  1. K2

    K2 Guest

    For an example of a Bayer design that uses no anti-aliasing, people
    should look at samples from the Kodak Pro 14n (nearly 14 megapixels).

    http://www.kodak.com/global/en/professional/products/cameras/dcsPro14n/dcsPro14nIndex.jhtml
    http://www.dpreview.com/reviews/kodakdcs14n/
    http://www.steves-digicams.com/2003_reviews/dcspro14n.html

    I think they look pretty darned good and definitely crisper than a
    standard blurry Bayer, though not quite a match for Foveon sharpness
    relative to the same screen area. Some subtle haze is added by the
    Bayer interpolation, in my opinion.

    The 14n has been around for many months but I thought it should be
    mentioned as a example of Bayer at its finest. The sensor is actually
    made by a company called FillFactory, and aside from not needing
    anti-aliasing, it's also CMOS-based.

    Apparently the 14n doesn't use anti-aliasing because the pixel array
    is so large that moire patterns are usually "out-resolved," but not
    entirely (see reviews). If it would only dispense with that RGB
    interpolation there'd be "no moire worries!"

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

  2. K2

    Giorgis Guest

    Pretty crappy troll effort. I would say that the 14n is plagued with
    problems.

    If only is had a penis as big as the Sigma.

    G
     
    Giorgis, Nov 28, 2003
    #2
    1. Advertisements

  3. K2

    K2 Guest

    I guess anything could be called a troll, but my post was trying to
    show a rare example of "pure" Bayer. Doctors recommend it, you know.

    There are indeed some big noise problems with the 14n and Kodak says
    it's best used for portraits with controlled lighting. Whether or not
    the noise is just a CMOS issue or something else remains to be seen.

    I have also stated that I'm no fan of Foveon's implementation in the
    Sigma SLRs, or of any digital SLR because of dust issues.

    K2
     
    K2, Nov 28, 2003
    #3
  4. I do see obvious Bayer aliasing in Kodak's samples. At 3000x4500
    pixels, I doubt it's a problem for normal viewing. Only a professional
    print shop has anything that can resolve 13.5 million full color pixels.
    That's 150 dpi even for a 20x30 inch poster.
     
    Kevin McMurtrie, Nov 28, 2003
    #4
  5. K2

    Mxsmanic Guest

    In all cases, increasing the number of pixels is preferable to
    anti-aliasing.

    In other words, it's better to have a sensor that has 14 megapixels and
    no anti-aliasing than a sensor with 6 megapixels and anti-aliasing.
    Anti-aliasing destroys image information; it is used only to hide
    certain artifacts of regular, low-frequency sampling that can appear in
    certain image configurations. It's not a substitute for higher
    resolution. And if your resolution is high enough, the aliasing is so
    slight that you don't need any special filter to compensate for it.
    See above. It's always better to increase resolution rather than mount
    an anti-aliasing filter. Anti-aliasing is the poor man's substitute for
    high resolution.
    There would still be aliasing effects, but they would be 3-4 times less
    obvious, and with high-resolution sensors, no anti-aliasing filter would
    be necessary.
     
    Mxsmanic, Nov 28, 2003
    #5
  6. K2

    Mike Guest

    Godwin, Do you agree that its strange that some so called photographers
    refer to the sex organs when talking about cameras. Giorgis is one seems
    have difficulty in recognising the differences. Do you think he was
    mistreated as a child by a photographer or perhaps he had his first orgasm
    using a throwaway camera and has been trying to capture the magic moment by
    talking dirty ever since.
    MikeS
     
    Mike, Nov 28, 2003
    #6
  7. SNIP
    Although I understand what you intend to say, the assumption not correct
    because aliasing mirrors the unresolvable frequencies to lower frequencies
    which are very easy to see.

    Try a simple experiment. Using e.g. this file:
    http://www.xs4all.nl/~bvdwolf/main/downloads/Rings.gif
    it's a 1000x1000 pixel 1.21MB file with concentric sinusoidal waves, with an
    increasing frequency towards the edges. View it at a 100% zoom setting in
    your photoeditor. Downsample this to e.g.10% of its original dimensions
    (100x100pixels), so your detail is 10 times smaller than your pixel
    resolution. Aliasing in its extreme is the result.

    You can use any resampling method you know of, Nearest Neighbor being the
    sharpest and the worst and Lanczos windowed Sinc more accurate. Only
    applying a little blur (similar to anti-aliasing or low-pass filtering)
    before downsizing can reduce the artifacts. Try a Gaussian blur radius of
    about 2 for the 10x reduction and compare. After resizing one should apply
    USM or a similar sharpening to restore the finest frequency contrast.

    I do agree that very small area aliasing is less likely to be visible
    because of the large pixel dimensions, which don't need as much enlargement.
    But large area aliasing will be painfully visible (try shooting regular
    pattern subjects like flat cloth, with or without herring bone patterns).
    And there is always detail smaller than the smallest sampling frequency
    available in a sensor, all it takes is distance and a sharp lens (a poor
    lens functions as a low-pass filter).

    Bart
     
    Bart van der Wolf, Nov 28, 2003
    #7
  8. K2

    Rafe B. Guest


    Bull. All sampled systems need anti-aliasing. It makes
    no difference what the sampling rate is. Any frequency
    component higher than 1/2 the sampling rate will cause
    aliasing. It makes no difference if it's a Foveon or Bayer
    sensor, and it makes no difference how many total pixels
    there are in the sensor, or what their spatial density is.

    Of course, at some point, the camera's lens will probably
    perform all the band limiting that's required -- ie., when
    the lens' resolving power itself guarantees the required
    f/2 bandwidth limit.


    rafe b.
    http://www.terrapinphoto.com
     
    Rafe B., Nov 28, 2003
    #8
  9. K2

    Mark M Guest

    PLONK
     
    Mark M, Nov 28, 2003
    #9
  10. K2

    JPS Guest

    In message <>,
    I looked at some of those samples, and they *do* have aliasing problems.
    I see color artifacts on single, in-focus hairs against a dark
    background, as well as stair-stepping in the more in-focus areas.
    --
     
    JPS, Nov 28, 2003
    #10
  11. K2

    Mxsmanic Guest

    If the sampling is far above what you can see, you don't need
    anti-aliasing.
    The higher the sampling rate, the more unlikely that becomes.
    Now you're getting the idea.
     
    Mxsmanic, Nov 28, 2003
    #11
  12. Although the 14n, with its 7.9 micron pixel pitch (or 127 lines/mm, or 63
    lp/mm native luminance resolution) still requires a poor lens to do that.
    IMHO a more efficient (only affects the highest frequencies) solution is a
    good lens + anti-aliasing filter.

    Bart
     
    Bart van der Wolf, Nov 28, 2003
    #12
  13. K2

    eawckyegcy Guest

    Note that a PNG version of the image is about 30% smaller than the
    GIF.

    Nice test image. Not surprisingly, Internet Exploder mangled it badly
    when it shrank the image. The result is difficult to look at;
    epileptics should leave the room. Adobe's PhotoSlop Elements (it came
    with the 10D - I don't use it) offers only NN, bilinear or bicubic.
    The latter is better, but it still has artifacts if the image isn't
    blurred prior to resample. ImageMagick does an excellent job:

    $ convert -geometry 100x100 Rings.gif Rings.png

    There isn't much difference even if you add a -blur prior to the
    -geometry. Curious boundary value effects after the blur bleed into
    the resampling though (for both IM and PS-E).

    Most interesting of all is how you can explore your own visual
    response by looking at the image.
     
    eawckyegcy, Nov 29, 2003
    #13
  14. K2

    K2 Guest

    No comment on that! But I can tell that many people in this group are
    extremely conservative about new technologies, Freudian motives or
    not.

    K2
     
    K2, Nov 29, 2003
    #14
  15. K2

    K2 Guest

    So, I guess it offends you that I'm showing examples of something
    non-Foveon that doesn't use anti-aliasing?

    How about less plonking and more looking (with your eyes) at the
    samples? Or are you afraid to learn anything new?

    K2
     
    K2, Nov 29, 2003
    #15
  16. K2

    K2 Guest

    What resolution is your monitor set to, what's the dot pitch and
    what's the size? If you answered that in another post, forgive me for
    repeating it.

    K2
     
    K2, Nov 29, 2003
    #16
  17. K2

    K2 Guest

    Amen. And the resolution setting of your monitor matters a lot, too.
    Those people are obsessed with aliasing to the point that I don't
    think they can enjoy the actual photos.

    K2
     
    K2, Nov 29, 2003
    #17
  18. K2

    K2 Guest

    What Bible did you pull that from? Why be so stuck on one idea?
    Or when excellent monitor resolution renders it moot anyway. Printing
    creates mechanical anti-aliasing as inks or dyes unavoidably bleed a
    little. For that reason I would rather send an aliased image than a
    pre-blurred one to a printer, since the printer will blur it enough.

    K2
     
    K2, Nov 29, 2003
    #18
  19. K2

    K2 Guest

    Whatever aliasing there is will be blurred away when printed on paper.
    I just don't see it as a big deal.

    K2
     
    K2, Nov 29, 2003
    #19
  20. K2

    K2 Guest

    Yes indeed.
    Ditto again. But the way you view an image (monitor resolution or
    printer resolution) is a big part of the equation.
    It's not I who disagrees on that point. Talk to these guys who claim
    "all" form of digital capture needs anti-aliasing or it's somehow
    violating an ethic.

    K2
     
    K2, 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.