JPEG 2000: which camera will support it? When?

Discussion in 'Digital Photography' started by John Faughnan, Jan 28, 2004.

  1. Because you probably didn't use appropriate JPEG compression settings.
    For example, JPEG-2000 uses arithmetic coding, so you must also use JPEG
    with arithmetic coding for comparision, otherwise you are comparing apples
    with oranges.
    Standard JPEG can also do 12 bits per pixel lossy and 16 bits per pixel lossless.
    RAW from a Bayer 'image'? Well, that is a good match, because both Bayer
    RAW and JPEG2000 are obsolete.

    Regards
    Guido
     
    Guido Vollbeding, Feb 2, 2004
    #21
    1. Advertisements

  2. I'm just using English with artistic latitude ;-).
    I have studied JPEG2000 enough to see that it is inferior.
    Most people don't understand JPEG, being lean and simple in comparison.
    It is not necessary to understand the cumbersome and inferior JPEG2000.
    Better learn to understand JPEG, that's my advice.
    In fact, I only recently discovered the major reason of JPEG's
    suitability - even the original JPEG authors, not to mention the
    JPEG2000 authors, didn't know the core JPEG property.

    Regards
    Guido
     
    Guido Vollbeding, Feb 2, 2004
    #22
    1. Advertisements

  3. John Faughnan

    Ron Hunter Guest

    I don't believe the word 'Bayer' appeared in his post. So you inserted
    it to make your point? This is not ethical debate tactics.
     
    Ron Hunter, Feb 2, 2004
    #23
  4. With some research you will find that the poster did not use the only
    available non-Bayer (or better non-mosaic) RAW camera (Sigma/Foveon).
    You will find that he used a Nikon D100.

    So yes, I made the point that using one obscure technology (capturing
    Bayer RAW) with another obscure technology (J2K compression) might be
    a good match. I recommend to avoid both.

    Regards
    Guido
     
    Guido Vollbeding, Feb 2, 2004
    #24
  5. John Faughnan

    Ron Hunter Guest

    RAW is RAW, it is the best the sensor can deliver, unprocessed. Seems
    like a good place to start if you want the best image possible from a
    given camera.
    Are you sure you don't see George when you look in the mirror?
     
    Ron Hunter, Feb 3, 2004
    #25
  6. RAW from a Bayer camera is far away from a natural photograph,
    missing 2/3s of the natural color spectrum!

    Regards
    Guido
     
    Guido Vollbeding, Feb 3, 2004
    #26
  7. John Faughnan

    Ron Hunter Guest

    If it is the best the sensor has to offer, it has to do.
    Note that it is modelled after the way the human (primate) eye works, so
    can we really SEE any difference, given the 'hardware' of the human eye?
    It's rather like spending money on a stereo to get flat response from
    20 to 22,000hz, when one can't hear past 7,500hz. Why entertain the dog?
     
    Ron Hunter, Feb 3, 2004
    #27
  8. The human eye is different and does NOT work like the Bayer sensor!
    My eye certainly doesn't like the Bayer look.

    Regards
    Guido
     
    Guido Vollbeding, Feb 3, 2004
    #28
  9. John Faughnan

    Lionel Guest

    It does, actually. Hit the library & read up on how the human retina
    works. It's far more similar to Bayer sensors work & totally unlike the
    way that Foveon sensors (or film, for that matter) work.
    You & George like what your sugar-daddies at Sigma tell you to like.
     
    Lionel, Feb 3, 2004
    #29
  10. Stay on-topic, please!
    Guido,

    Turning it round, then, considering that the majority of cameras today
    _do_ use Bayer mosaics, how does that affect what compression should be
    used? I mean, what compression would be more suitable than JPEG if a
    certain file size is required, or can one do a reasonable job simply by
    tuning the parameters of JPEG such as chroma sub-sampling?

    Cheers,
    David
     
    David J Taylor, Feb 3, 2004
    #30
  11. Turning it round, then, considering that the majority of cameras today
    David,

    JPEG and other image compression and image processing schemes are based
    upon the basic digital imaging model. A key property of this digital
    imaging model is the full-color-spectrum-per-pixel assumption.
    Noticeably violating this basic assumption in the case of Bayer sensing
    renders all further results from the usual model useless.
    If I would see any opportunity to produce useful results from a Bayer
    capture, I would let you know, honestly.
    But my perception is that there is no such opportunity, so whatever you
    do to compress/process a Bayer 'image', you will not get something near
    a natural photograph experience.

    If you are in the business to define an appropriate representation of
    natural photographs in a digital form, you must start from the closest
    approximation of such image. The usual true-color digital image is such
    close approximation, but the Bayer 'image' definitely is not, because it
    already lost a considerable amount of picture information. No digital
    image compression expert would come to the idea to reduce the amount of
    picture data in the Bayer pattern way - the Bayer approach is nothing
    more than a technical artifact in times when people tried to capture
    color images with monochrome image sensors. Nice try, but now we see
    that it failed, and we need native (true-)color sensors.

    Now here is a rough description about my recent JPEG discovery in this
    relation:
    The fundamental property of lossy image compression is the similarity
    of different resolutions of the same image. "Lossy" compression means
    that we assign *the same* output representation to *multiple*, *similar*
    input representations. The basic similarity relation for images is
    resolution, or scale, invariance: If we see the same image in different
    resolutions (scales, sizes), or the same subject from different distances,
    we talk about *the same* image (or subject).
    Resolution is, beside the true-color-per-pixel property, one part of the
    basic digital imaging model. JPEG is the optimal algorithm for lossy
    image compression, because the DCT (Discrete Cosine Transform - the core
    of JPEG) provides the best resolution separation property for digital
    images. Wavelet transforms, as used in JPEG2000, for example, do *not*
    provide such optimal resolution separation. See also chapter 4 of my
    paper at http://jpegclub.org/temp/ .

    Regards
    Guido
     
    Guido Vollbeding, Feb 3, 2004
    #31
  12. David,
    Whilst I hear what you say, I don't perceive it that way.

    To me, JPEG can comprise a high-resolution monochrome channel, together
    with lower resolution colour channels. Under those circumstances, it
    seems to me idea for the Bayer sensor with its differing luminance and
    chrominance resolutions.

    Cheers,
    David
     
    David J Taylor, Feb 3, 2004
    #32
  13. Where does the Bayer sensor separates or differs luminance and
    chrominance resolution?
    I constantly hear this argument, but I can't perceive it that way ;-).
    Look at the Bayer sensor or RAW format, please! Where do you have a
    proper luminance information there which would allow for capturing a
    "high-resolution monochrome channel"? It is just not there - all
    what you have are the Red, Green, or Blue elements - no proper
    luminance anywhere.

    So while JPEG can comprise a high-resolution monochrome channel,
    the Bayer scheme *prevents* the capture of a high-resolution
    monochrome channel. You cannot take away the color masks
    from the Bayer sensor even if you wanted just a B/W capture.

    Regards
    Guido
     
    Guido Vollbeding, Feb 3, 2004
    #33
  14. []
    Yes, I see what you mean - at best it's an approximation to "luminance",
    with some added noise.... It's amazing how well it works, then!

    Cheers,
    David
     
    David J Taylor, Feb 3, 2004
    #34
  15. The JPEG luminance formula is as follows:

    Y = 0.299 * R + 0.587 * G + 0.114 * B

    If you have just *one* of the RGB elements as in the Bayer case,
    you can see how well this "approximation to luminance" works, then...

    Regards
    Guido
     
    Guido Vollbeding, Feb 3, 2004
    #35
  16. The JPEG luminance formula is as follows:
    BTW, from this formula you can see why the Bayer pattern contains 2
    Green elements versus 1 Red and 1 Blue: Green contributes the most
    part to the luminance. But there's still a lot missing...

    Regards
    Guido
     
    Guido Vollbeding, Feb 3, 2004
    #36
  17. []
    Yes, if one considers the individual pixels. With the right anti-alias
    filtering, you can derive a good RGB from a quad of four, though. Then
    you could apply standard JPEG techniques.

    Cheers,
    David
     
    David J Taylor, Feb 3, 2004
    #37
  18. That formula has nothing to do with Bayer: it reflects human visual
    perception and applies to any RGB image. And talking about Bayer in terms of
    quads of pixels is simply wrong.

    The two of you have gotten yourselves confused.

    The problem at hand (compressing RAW images) should be handled by
    _independently_ and losslessly compressing (a) the set of R measurements as
    a monochrome image, (b) the set of G measurements as a monochrome image, and
    (c) the set of B measurements as a monochrome image. That set of three
    compressed images can then be independently reconstructed so that one can
    then apply better (more computationally intesive) Bayer demosaicing
    algorithms outside the camera.

    So the relevant question for the RAW problem is what are the best lossless
    monochrome compression techniques. Since RAW formats are proprietary, this
    has nothing to do with JPEG vs. JPEG2000.

    If you do your Bayer demosaicing in camera, then you've got an RGB matrix to
    which either JPEG or JPWG2000 can be applied, and which works better is
    going to be independent of Bayer demosaicing.

    David J. Littleboy
    Tokyo, Japan
     
    David J. Littleboy, Feb 3, 2004
    #38
  19. []
    David, thanks for your input. I must confess to thinking more about the
    composite RGB image, as I don't use RAW myself. I can see Guido's point,
    though, as JPEG typically implies separate luminance and chrominance
    sampling, and that perhaps the statistics of the luminance may differ
    according to the sensor type.

    Cheers,
    David
     
    David J Taylor, Feb 3, 2004
    #39
  20. But you will at some point, most likely. Not having to commit to an
    Snore. It'll mean that images from a 3-color system compress less well, but
    other than that, it's irrelevant. Guido's "point" isn't a point at all: it's
    a stupid, technically incorrect, rant. Bayer is flipping amazing: you get as
    much luminance information as can be correctly rendered in a matrix of that
    size (without aliasing) and you get as much color as the human eye can
    detect. There isn't anything missing that you can see or use.

    And anyway, since the only cameras that actually measure 3 colors everywhere
    are dogs, they're irrelevant.

    David J. Littleboy
    Tokyo, Japan
     
    David J. Littleboy, Feb 3, 2004
    #40
    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.