maximum dynamic range using RAW

Discussion in 'Digital Photography' started by Frank van der Pol, Dec 21, 2003.

  1. - as the ADCs are typically 10, 12 or 14-bit, and the eye can only see
    7
    Not having measured the output from a blue channel under artificial light,
    I don't know if you are correct or not. Is it really as low as 1/8 of its
    value in daylight conditions? Oh, if only we stored the values as logs!
    All I'm saying is that noise, rather than quantisation error, may be the
    most important factor.

    Additionally, as green contributes more to the brightness of the picture
    compared to red or blue, there is an argument for using one more bit for
    the green channel than the other two, i.e. more closely matching the
    different contrast sensitivity of the eye to brightness and colour. Of
    course, in practice prviding three of the same is the simplest solution.

    CHeers,
    David
     
    David J Taylor, Dec 22, 2003
    #21
    1. Advertisements

  2. Frank van der Pol

    Andrew Guest

    Dynamic range is not measured in bits. Are you saying that given the dynamic
    range of the the eye, whatever that may be, that the eye can only resolve
    128 distinct gradations when adjusted to particular conditions? I can tell
    you that when I look at a test image on my computer monitor of 256 levels
    of grey, from black to white, I have no trouble at all distinguising between
    each and every grey level. And I'm sure that my eye can resolve levels darker
    and lighter than my monitor can generate without readjusting. Are you saying
    that the eye does a mini-adjustment when it moves from a lighter to a darker
    portion of the same photograph?

    Andrew
     
    Andrew, Dec 22, 2003
    #22
    1. Advertisements

  3. That is the same as the audio equivalent of making a
    Audio is another of those subjective areas. When people started claiming
    they could hear the difference with gold-plated main connectors..... well!
    However, do you really think it's as bad as -40dB - the difference between
    daylight and artifical? I guess the transmission curve of a conversion
    filter might tell us!

    Cheers,
    David
     
    David J Taylor, Dec 22, 2003
    #23
  4. Frank van der Pol

    Chris Brown Guest

    For starters, you need to "expose to the right", i.e. the brightest thing in
    your farme should *just* be approaching blowout. That way, you maximise the
    signal to noise ratio, because the noise is mostly in the low-order bits.
    Indeed. I've often wondered what the neutral colour balance of my 10D is.
    Clearly, the further away you have to go to white-balance the image, the
    noisier the end result is going to be (because you'll be boosting one
    channel dispropotrionately). One thing I've noticed in particular is that an
    exposure using 800 or 1600 ISO with "daylight" as its final white balance is
    much less noisy than an equivalent exposure set to "tungsten" or
    "flourescent", so I suspect the neutral white balance is pretty close to
    daylight.
    It would be nice if there was a "camera" white balance setting. Not seen any
    raw conversion software that supports this, sadly.

    If you want to experiment, however, hacking the open source convertor,
    dcraw, so that it doesn't do any colourspace conversion would be a good
    starting point.
     
    Chris Brown, Dec 22, 2003
    #24
  5. Frank van der Pol

    Andrew Guest

    Dynamic range is not measured in bits. Giving you the benefit of the doubt
    that what you mean is 7 bits of gradation withing a given dynamic range,
    the burden is on you to give a citation for the 7-bit idea because this
    is counter to every informed opinion on the subject that I've read in this
    and every other forum.

    Andrew
     
    Andrew, Dec 22, 2003
    #25
  6. David J Taylor schreef in bericht
    Correct. But that is achieved by correctly exposing the image, while
    at the same time making sure none of the cells gets a n amount of
    light that results in a value that can not be converted by the A/D
    converter.

    But a correct exposure also means that white light should result in
    getting an equal amount of signal from each set of colored cells in
    the bayer pattern.

    White should therefore result in a maximum (equal) charge in each of
    the CCD cells. Question is basically: What is the color white that
    will result in an equal analog charge value at the analog side of the
    A/D converter i.e. will result in equal digital values at the digital
    side of the A/D converter?
    Controlled situations like products in a studio actually. So the
    absolute amount of light is not a real issue.

    Frank
     
    Frank van der Pol, Dec 22, 2003
    #26
  7. But a correct exposure also means that white light should result in
    I don't think that you will get the same charge. All the cells are [most
    likely] panchromatic, and therefore you will get differing amounts of
    charge from the R G B (or C M Y) filtered cells. Again, I suggest looking
    at the spectral response of a "daylight to artificial light" filter to see
    what the difference might be. I've looked quickly, but can't find any
    plots. Filter factors seem to be between 0.5 and 1 stop.

    []
    So much light there I would have thought dynamic range ceased to be an
    issue if the exposure was correct....

    Cheers,
    David
     
    David J Taylor, Dec 22, 2003
    #27
  8. David J Taylor schreef in bericht
    The difference is that having solid gold railroad tracks feeding your
    speakers is done with the intention not to change anything to the
    original signal.

    The goal of the imaging equivalent is to get as much signal in order
    to have enough quality to be able to make
    I't just an assumption. I can't measure this. But I think that to get
    a good view of what the actual result may be, you should not look at
    the light only, since what really counts is the result when making an
    exposure of a certain object.
    The difference is always in the extremes. One won't notice any quality
    difference probably when shooting a neutral gray subject on a white
    background in either 6000K or 3200 K. But when you shoot say a red
    subject on a blue background using 3200 K tungsten, this might well
    result in severe clipping of the red colored cells in the bayer
    pattern while the blue colored cells may show numbers close to zero.
    But that only says something about the percentual differences, not
    about absolute values that reach the CCD cells using a real
    image/subject that contains colored objects.


    Frank
     
    Frank van der Pol, Dec 22, 2003
    #28
  9. Dynamic range is not measured in bits. Giving you the benefit of the
    doubt
    Bits is a convenient unit to use for digital rquipment, though.

    I believe I am talking dynamic range within a single black-and-white
    image, viewed at normal viewing distance on a CRT display, not walking
    right up to the display or using a lens to magnify parts of the image.

    View such an image with an 8-bit quantisation (typical of most display
    systems). View such an image with 7-bit quantisation (at least typical of
    the errors introduced by low-compression JPEG, or the levels of dither
    used in stenography).

    Can you see the difference?

    Cheers,
    David
     
    David J Taylor, Dec 22, 2003
    #29
  10. Dynamic range is not measured in bits. Are you saying that given the
    dynamic
    Yes, I am talking about viewing an overall scene. Can you reliably
    distinguish between a scene with 256 levels and one with 128 levels? Not
    a testcard. The eye does ajust when looking away from a brightly
    displaying monitor to a dark portion of the room - I just tried that!

    Cheers,
    David
     
    David J Taylor, Dec 22, 2003
    #30
  11. Frank van der Pol schreef in bericht
    ....adjustments without noticable quality decrease.

    ;-)

    Frank
     
    Frank van der Pol, Dec 22, 2003
    #31
  12. Audio is another of those subjective areas. When people started
    claiming
    That was meant to read "mains connectors" - in the UK that means power
    connectors. In short, much os what was written about audio in the "Hi-Fi"
    magazines was rubbish.

    []
    It is crying out for measurement. Don't these expensive cameras that need
    RAW files tell you what the actual values are? I still think you are
    being pessimistic.

    Cheers,
    David
     
    David J Taylor, Dec 22, 2003
    #32
  13. David J Taylor schreef in bericht
    Ooops. You are right. But then the maximum electrical value at the
    output of the CCD will be (should be) made equal using amplifiers in
    order to make use of the maximum range of the A/D converter and the
    maximum of available values in the RAW file.
    But correct exposure is correct for each of the channels. And that is
    achieved by using the correct color of the light.

    Frank
     
    Frank van der Pol, Dec 22, 2003
    #33
  14. The goal of the imaging equivalent is to get as much signal in order
    Not sure I agree 100%. I don't have a lot of time for extensive
    post-processing on a large number of pictures. I would therefore put
    getting it 99% correct first time above the ability to fix-up images
    afterwards. Yes, of course some images would benefit from
    post-processing, and then I do with my camera saved more bits!

    Cheers,
    David
     
    David J Taylor, Dec 22, 2003
    #34
  15. Frank van der Pol

    Chris Brown Guest

    That's tricky, but might be doable for certain scenes.

    256 levels and 64 levels is easy, unless you resort to error-diffusion (AKA
    "dithering").
     
    Chris Brown, Dec 22, 2003
    #35
  16. David J Taylor schreef in bericht
    But you seem to forget that the conversion from RAW to tif and in the
    process correcting the color balance so that in any given neutral gray
    or white area the value for R = the value for G = the value for B in
    the tif image _is_ to some degree the same as fixing up an image
    afterwards.

    When in a 6000K white area in a subject the digital value of pixels
    that have a blue filter in the bayer pattern only have a value of 80
    in the raw file, while the pixels that have a red filter already have
    a value of 250, you will be more or less unknowingly fix this shortage
    of color resolution in the blue channel while you convert from RAW to
    RGB.

    Frank
     
    Frank van der Pol, Dec 22, 2003
    #36
  17. I don't think that you will get the same charge. All the cells are
    [most
    Well, you would have to set the maximum values so that any one scene pixel
    would not clip, and that _all_ scene pixels would not clip. I think the
    only simple way of doing that (in the absence of flourescent colours)
    would be to assume that white = maximum in R, G and B. Then you need a
    margin for specular highlights, so set R G and B at 80% of the FSD? (full
    scale deflection). Or do you allow for much greater specular highlights
    and set "white" as being 50% FSD? I don't know.
    May be. But I still don't think that for most pictures, under studio
    lighting conditions, there would be a lot to gain from adjusting the
    relative gains of the R G B channels by (what I suspect would be) just a
    few percent.

    Cheers,
    David
     
    David J Taylor, Dec 22, 2003
    #37
  18. But you seem to forget that the conversion from RAW to tif and in the
    Agreed, there are gain adjustments being made, and these should be made
    with at least the same precision as the data, if not a bit or two better.
    I.e. if the RAW dat is 10 bit, then look to process to at least 12 bits.
    If you don't do this, then you will end up with missing levels from the
    10-bit data.
    But I would hope that the values in the RAW file were 1023 or 4095 for the
    maximum values! In the 10-bit case, blue in your example would be stored
    as 320 and read as 1000, so it has a resonable margin for processing.

    If your RAW file is only 8-bit precision, then I think it is a complete
    waste of time.

    Cheers,
    David
     
    David J Taylor, Dec 22, 2003
    #38
  19. Yes, I am talking about viewing an overall scene. Can you reliably
    Oh, yes. Even I wouldn't push a 6-bits is enough claim, although given a
    small amount of noise in the source image.....

    Cheers,
    David
     
    David J Taylor, Dec 22, 2003
    #39
  20. David J Taylor schreef in bericht
    But if the actual data within the 10 bit space is no more than 8 bits
    because the level of one channel does not get beyound that value as a
    result of a mismatch of light color vs sensitivity of the CCD, you
    will never get more than 256 discrete values within the full 10 bit
    scale of that channel.
    I agree.

    Frank
     
    Frank van der Pol, Dec 22, 2003
    #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.