Aspect ratio - why not exactly 3:2

Discussion in 'Digital Photography' started by David J Taylor, Jan 31, 2004.

  1. It seems that the Nikon D100 and D70 produce 3,008 x 2,000-pixel images.
    Why not a more precise 3:2? I am guessing that pixel count has to be a
    multiple of 16 for processing into JPEG, so that the horizontal count must
    be a multiple of 48 and the vertical a multiple of 32. 2976 * 1984
    doesn't sound as good, does it? I guess the CCD doesn't support Canon's
    3072 x 2048.

    So is the 3008 just to avoid a figure like 29xx? Marketing in other
    words?
    Does the lack of precise 3:2 ever matter?

    Just some random thoughts....

    David
     
    David J Taylor, Jan 31, 2004
    #1
    1. Advertisements

  2. David J Taylor

    Mark Herring Guest

    Why???? I can make an image in PS any size I want and still save as a
    JPEG
    I suspect that one factor is the number of pixels dedicated to
    calibration---ie pixels that exist in the CCD (or CMOS sensor), but do
    not appear in the image files.
    **************************
    Mark Herring, Pasadena, Calif.
    Private e-mail: Just say no to "No".
     
    Mark Herring, Jan 31, 2004
    #2
    1. Advertisements

  3. Why???? I can make an image in PS any size I want and still save as a
    Because the JPEG blocks are 8 x 8, or 16 x 16, or whatever, and therefore
    might produce fewer artefacts if the images are multiples of these sizes.

    []
    Yes, of course, Mark. It just struck me as a slightly "odd" number and I
    had wondered why.

    Cheers,
    David
     
    David J Taylor, Jan 31, 2004
    #3
  4. David J Taylor

    Ron Hunter Guest

    I guess if 8 pixels trimmed of the side will mess up your composition,
    it might. I don't really think I would ever lose any sleep over it...
    Grin.
     
    Ron Hunter, Jan 31, 2004
    #4
  5. David J Taylor

    Mark Herring Guest

    I don't think so. Any compression algorithm has provisions for what
    to do at the edges of the scene. If it is working in blocks, and gets
    an "incomplete block", then MAYBE there will be some different
    artifact in that area only.
    there are many things in electronic design that make no sense when you
    look at the finished product. (Sometimes they don't even make sense
    when you unravel it---there may be just the engineer's whims)
    **************************
    Mark Herring, Pasadena, Calif.
    Private e-mail: Just say no to "No".
     
    Mark Herring, Jan 31, 2004
    #5
  6. David J Taylor

    Don Stauffer Guest

    I print a lot of 8 x 10. I'd prefer a precise 5:4 ratio.
     
    Don Stauffer, Jan 31, 2004
    #6
  7. There are times when I think a square aspect ratio would make things a lot
    easier!

    Perhaps not for the landscape enthusiast, though!

    What I have found is that the combination of 4:3 and 3:2 is actually quite
    handy, combines general purpose with landscape. Perhaps even 1:1 and 2:1
    would be an even better pairing?

    Cheers,
    David
     
    David J Taylor, Jan 31, 2004
    #7
  8. David J Taylor

    JPS Guest

    In message <x3LSb.2270$>,
    Might make the firmware smaller and faster, too, if it doesn't have to
    deal with special cases.
    --
     
    JPS, Jan 31, 2004
    #8
  9. JPEG compresses 8x8 blocks--it doesn't require multiples of 32 or 48.

    3008 does happen to be divisible by 16, while 3000 wouldn't be. This is a
    major convenience for Windows XP users, because XP's built-in lossless
    rotation requires both dimensions to be a multiple of 16 pixels.

    For example, my Canon A80's highest resolution is 2272 x 1704. 1704 is not
    divisible by 16, so XP will not do a lossless rotation on these photos. (I
    solved this problem by using exifiron which rotates automatically based on
    the orientation sensor, and does a lossless rotation at this resolution.)

    I don't know if the XP rotation issue is why Nikon chose that size, but it
    certainly works to the benefit of XP users.

    -Mike
     
    Michael Geary, Jan 31, 2004
    #9
  10. If you factor in the 3:2 aspect ratio requirement, then the picture block
    size becomes 24 x 16 (or 48 x 32 if you need a 16 x 16 jpeg block size).
    That's where those numbers come from.
    Thanks, Mike. I actually use PIE Studio to lossless rotation rather than
    XP, so I've not compared XP and PIE Studio as to their JPEG size
    requirements. I had a sneaky feeling choosing a number just greater than
    3000 might have been marketing! I expect Nikon engineering have some
    choice in exactly how many of the available sensor pixels they choose to
    use for active pixels.

    Cheers,
    David
     
    David J Taylor, Jan 31, 2004
    #10
  11. If the dimensions are a multiple of 16, then only full JPEG blocks are
    needed. This simplifies the encoding hardware/firmware in the camera
    (with no partial blocks, no need for special cases in the firmware).

    It also means that you can losslessly rotate the image without losing
    any of it. JPEG decoders are prepared for images that are not a
    multiple of 16 in size, but they assume the "surplus" pixels are located
    on the right and bottom edges. That's no longer true after lossless
    rotation, so the programs that do the rotation have to trim the image
    slightly or something equivalent.

    But if the image is a multiple of 16 in both dimensions, everything
    works as expected.

    Dave
     
    Dave Martindale, Jan 31, 2004
    #11
  12. In one sense, a square image sensor makes the best use of the lens
    coverage. Then you could crop either a vertical or horizontal image
    from it after the fact.

    Going even further, the best sensor would be circular with a size large
    enough to capture the full useful image circle of the lens. This would
    have to be written to a square output file whose corners were black, but
    extra black pixels in a JPEG or RAW file (at least Canon raw) don't cost
    hardly anything. This would allow you to extract either horizontal or
    vertical images with a larger field of view than you'd get from the same
    lens with a square sensor where the corners had to be within the image
    circle.

    Trouble is, you have to crop every image before printing,
    which the family snapshot enthusiast might not appreciate.

    Dave
     
    Dave Martindale, Jan 31, 2004
    #12
  13. But if 2:1 chroma downsampling is used, the source block size is really
    16x16.

    Dave
     
    Dave Martindale, Jan 31, 2004
    #13
  14. If you insist on an exact 3:2 or 4:3 aspect ratio. But film cameras
    were never that exact in the dimensions of the film gate.

    For example, suppose some camera produced an image that's 3056x2048.
    Both dimensions are multiples of 16 for easy JPEG handling. The aspect
    ratio is 3:2 within 0.5%. I'll bet most film cameras are further off
    than that.

    Dave
     
    Dave Martindale, Jan 31, 2004
    #14
    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.