Dynamic range of digital and film: new data

Discussion in 'Digital Cameras' started by Roger N. Clark (change username to rnclark), Nov 7, 2004.

  1. Hi.

    I've been running tests and have some interesting new plots
    of film and digital dynamic range. This is the first of about
    10 graphs on this page, but I though I would share it for comments
    while I built the rest of the page. See:

    http://clarkvision.com/imagedetail/dynamicrange2

    The figure on the pages shows the transfer function of a digital camera
    (a Canon 1D Mark II) compared to print file (Kodak Gold 200), slide
    film (Fujichrome Velvia), and the relative response of the
    Human eye (note the human eye has a much greater dynamic range).
    What this plot shows is that the digital camera response function
    is similar to print film, but even lower in contrast. The
    response of both the digital camera and print film shows lower
    contrast than apparent to the human eye (the steeper the rise,
    the slope, the greater the contrast). Fujichrome Velvia has
    the highest contrast of the 4 systems shown. But also of interest
    is the noise. Note that the digital camera points follow a nice
    smooth line. The Fujichrome Velvia points follow a trend with
    scatter of individual points several times larger than the digital
    camera. The print film shows the widest scatter and therefore
    has the highest noise. Also of interest is where each curve
    flattens out in the lower left corner. Note the Fujichrome
    Velvia slide film begins to flatten at about 3000 DN on the
    Scene Intensity, and really flattens out just above 1000 DN.
    The print film flattens out (and also becomes excessively noisy)
    below about 900 DN. But the digital camera keeps going to the
    bottom end of the data at 300 DN. To be added: lower values:
    the digital camera continues providing good data down to a
    few DN! This shows the Canon 1D Mark II has a much higher
    dynamic range than either Fujichrome Velvia slide film and
    Kodak Gold 200 print film. Kodak Gold 200, in this test,
    showed 7 stops of information, Fujichrome Velvia 5 stops,
    and the Canon 1D Mark II, over 11 stops of information!

    Roger
     
    Roger N. Clark (change username to rnclark), Nov 7, 2004
    #1
    1. Advertisements

  2. Flipping excellent work! Really lovely.

    David J. Littleboy
    Tokyo, Japan
     
    David J. Littleboy, Nov 7, 2004
    #2
    1. Advertisements

  3. Roger,

    This is looking very interesting, and I await the results of the rest of
    the work. Could I suggest, though, that the plots be on a log-log rather
    than a linear-log basis? I guess that this would make the eye a
    straight-line plot., and probably emphasise the dark-region noise of the
    Kodak 200 rather. Different contrasts would simply be different slopes of
    line rather than forcing you to guess curve shapes...

    What do you think?

    Cheers,
    David
     
    David J Taylor, Nov 7, 2004
    #3
  4. Yes, I'll include log-log. Both plots are useful and separate the
    curves in different regions.

    Roger
     
    Roger N. Clark (change username to rnclark), Nov 7, 2004
    #4
  5. Roger N. Clark (change username to rnclark) wrote:
    []
    Thanks, Roger. David
     
    David J Taylor, Nov 7, 2004
    #5
  6. Roger N. Clark (change username to rnclark)

    bob Guest

    I can't wait to see the rest.

    It seems to indicate that there might not be a lot of room for improvement
    in top end CCDs.

    Bob
     
    bob, Nov 7, 2004
    #6
  7. Roger N. Clark (change username to rnclark)

    Ken Tough Guest

    Greater on the low end, but does the human eye give useful
    response at the high end (above where the graph ends and
    it goes exponential?)
    Except at low intensities, though? The slide film is virtually
    flat below scene intensity of 3000, meaning it really isn't
    producing anything useful. It's towards the higher end where
    slide gets better. (as you point out below)

    Nice science!
     
    Ken Tough, Nov 9, 2004
    #7
  8. Roger N. Clark (change username to rnclark)

    Clyde Guest

    It helps that the eye isn't viewing just one picture. It's "taking" a
    whole bunch of pictures that the mind is splicing together. The human
    eye only sees about 1 degree in sharp, clear focus. The rest is slightly
    to very blurry.

    Heck, we have a blind spot in our vision that is big enough to cover 6
    full moons stacked on top of each other. We don't even see that, even
    though it is in our area of "vision".

    All this means that the human eye is adjusting more than it's focus as
    it scans. It is also adjusting its exposure. That is a huge advantage in
    capturing dynamic range.

    Actually you can do the same thing with your digital camera too. Put it
    on a tripod, make sure nothing moves, shoot several shots at the
    extremes of exposure needed (and a few in between), then pick the parts
    you want to use in the final picture. This can give you a VERY large
    dynamic range. Alas, it doesn't work well for moving subjects.

    Clyde
     
    Clyde, Nov 10, 2004
    #8
  9. Roger N. Clark (change username to rnclark)

    JPS Guest

    In message <>,
    Your figures are impossible. You can not have 12-bit image data that is
    4096; perhaps you mean 4095. EWven if you are measuring 4095, you can't
    possibly measure 4 also. There isn't that much dynamic range, as there
    is a DC component to sensor noise that makes the lowest digitized 12-bit
    values in the hundreds.
    --
     
    JPS, Nov 10, 2004
    #9
  10. In checking data on my 10D and 1D Mark II, the lowest DNs are in the
    few range. For example, in a 1DII test image with a dark shadow
    designed to be really close to black, I see values with a mean of 4.3
    (0 - 4095 scale) up to 4042 (no part of the image was saturated).
    These are output in linear mode. These same pixels, when output
    with the default gamma in the canon converter, gives values
    for the dark shadow of 43 up to 4042 for the brightest.
    Perhaps you are looking at raw data converted with the standard
    curve?

    Roger
     
    Roger N. Clark (change username to rnclark), Nov 11, 2004
    #10
  11. Roger N. Clark (change username to rnclark)

    JPS Guest

    In message <>,
    I'm talking about RAW data with no curve at all. 12-bit linear, right
    out of the camera.

    Almost every program that gives you a "linear TIF" subtracts the
    blackpoint first, and then re-scales the data, before giving it to you.
    --
     
    JPS, Nov 11, 2004
    #11
  12. I think you are confusing "black point" with noise. Many systems
    put a DC bias in the signal so the A/D converter does not try and
    digitize a negative voltage (e.g. due to noise). That bias
    is not necessarily a noisly signal, and since it is artificial,
    you do want to subtract it off. So what is your point?

    Roger
     
    Roger N. Clark (change username to rnclark), Nov 11, 2004
    #12
  13. Roger N. Clark (change username to rnclark)

    JPS Guest

    In message <>,
    I don't think that any of the converters that offer a demosaiced "linear
    TIFF" are going to show you accurate information about the RAW data.

    The two options I am aware of are:

    1) Take a non-compressed .DNG conversion, and load it as .raw in
    photoshop (the old kind of RAW, not the digital camera RAW).
    Unfortunately, Adobe's DNG converter does not scale the values, so the
    data gets posterized when it gets loaded into Photoshop's pseudo-16-bit
    mode (it's really 15-bit plus one extra value of 32768). I am going to
    write a small console program that doubles the data in the DNG file, and
    strips the headers, if I get a chance soon. I am also going to write a
    version that adds multiple exposures together. 16 exposures added
    together will have 4x the signal-to-noise ratio, and 16x the
    levels-resolution in the darkest shadows (15x in photoshop's 15-bit).

    2) Convert with DCRAW in "document mode", and comment out the RGB
    conversion, and the color scaling. This gives you the RAW data, scaled
    better than Adobe's DNG, but unfortunately, DCRAW seems to clip
    highlights (at least on the 10D it does), you get 0 to 60 in
    0-to-255-space, where you should be getting 0 to 63. I don't know where
    this happens in the code, and I am not experienced enough at reading
    other people's code to figure out why. DCRAW has a lot of useful info
    in it, but the coding is a bit unwieldy, IMO. The order of the code in
    the source is totally random, and real code flow jumps all around the
    source. Lots of useful comments in some spots, and no comments at all
    in others.
    --
     
    JPS, Nov 12, 2004
    #13
  14. Roger N. Clark (change username to rnclark)

    JPS Guest

    In message <>,
    My point is that regardless of the source of this offset, it restricts
    the range of values in the RAW data, and when you are looking at values
    from a linear tiff, they are stretched, and demosiaced, and do not give
    you a good look at the real data.

    Quoting values of 4 and 4096 (or 4095), indicates that you are not
    looking at RAW values.
    --
     
    JPS, Nov 12, 2004
    #14
  15. Roger N. Clark (change username to rnclark)

    JPS Guest

    In message <>,
    It really matters to know exactly what it is you're talking about.
    --
     
    JPS, Nov 12, 2004
    #15
  16. I disagree. If you were studying the characteristics of the
    electronics, then true raw values would be relevant. But we
    are not concerned with the specifics of each electronic component,
    only what the SYSTEM delivers in image data. So, having the
    "black point" subtracted is what we want to evaluate, because
    that data is the image data that wee see. If subtracting the
    black point changes the total dynamic range and increases
    the noise, that is exactly what we want to measure because
    that is the data that effects the image, not something
    before that processing step. And in any case, a couple hundred
    DN offset has only a small percentage affect on the dynamic
    range, and in my measurements, though I believe pretty good, would
    be near the error in measurements.

    Roger
     
    Roger N. Clark (change username to rnclark), Nov 13, 2004
    #16
  17. You might take a look at the image processing program ImagesPlus
    ( http://www.mlunsold.com/ ) It works in true 16 bit, adds images
    together, and can register them before adding. It also has a
    lot of interesting filters and is widely used in the astronomy
    community. (I have no affiliation or monetary interest with mlunsold.com).
    It is what I use for all my camera statistics.
    Astronomers commonly add dozens to hundreds of images together
    to increase signal to noise. For example, see:
    http://clarkvision.com/astro/saturn.03.02.2004
    I added 99 frames of Saturn together, taken with a telephoto
    lens and applied filters to increase spatial resolution.

    Roger
     
    Roger N. Clark (change username to rnclark), Nov 13, 2004
    #17
  18. Roger N. Clark (change username to rnclark)

    JPS Guest

    In message <>,
    I downloaded the demo, but the demo seemed to be restricted from doing
    anything I was interested in trying, IIRC.

    Does its stacking feature automatically rotate and offset individual
    images, and does it use sub-pixel placement, outputting a higher-res
    image?
    --
     
    JPS, Nov 14, 2004
    #18
  19. I'm not sure about the sub-pixel placement. You could
    always enlarge the image first (it does have batch
    resize abilities, which is qhat I do). Send Mike (the author
    of the prog) an email. He is usually pretty good about responding.
    He adds features based on user requests too. I have done
    planet registration as well as astrophoto shift and rotate
    registration of star fields. You select one or two
    areas for it to match to and it does some sort of autocorrelation
    to do the alignment. It can also grade the quality of
    multiple images and add them in a weighted average.

    Roger
     
    Roger N. Clark (change username to rnclark), Nov 14, 2004
    #19
    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.