ISO and actual sensitivity in DSLR's (D70, *istD, 20D, S3...)

Discussion in 'Digital Photography' started by Alan Browne, Mar 25, 2005.

  1. Alan Browne

    JPS Guest

    In message <>,
    Actually, black is almost *exactly* 118 out of 4095 in a 20D RAW file at
    ISO 100, with a short exposure at room temperature.
    --
     
    JPS, Mar 29, 2005
    1. Advertisements

  2. Alan Browne

    JPS Guest

    In message <>,
    Nope. The numbers, as reported by the Info tool, change with color
    management changes.
    --
     
    JPS, Mar 29, 2005
    1. Advertisements

  3. Alan Browne

    JPS Guest

    In message <d29cl5$dnq$>,
    What are you talking about? What results? The fact that ACR or PSE3
    and the JPEG have the same value? They both make the same assumptions,
    but that does not tell us where this mid-grey lies in the RAW data. It
    could still be 2.2 stops below the RAW clipping point, or 5 stops, in
    the green channel. The other channels are different (and usually vary
    by as much as a stop). It's arbitrary, but known by both converters.

    This is a fact:

    If two cameras claimed a lowest native ISO of 100, and both had mid-grey
    or "mid-green", which would map to 118 in the output, one could be doing
    it with a RAW green value twice as high (after blackpoint subtraction)
    as the other. One would have twice as much headroom as the other, and
    if that other could legitimately be called "ISO 100" based on
    sensitivity, the one with the extra stop of headroom could be called
    "ISO 50", if you just moved the mid-green a stop higher. In fact, this
    is exactly what you can do on some cameras when you shoot RAW instead of
    JPEG; you get a usable stop of extra highlights, and you could set the
    EC to +1 and say you're shooting at half the ISO. Both my 10D and 20D
    can take shots with non-specular highlights, and work off of an external
    meter set to ISO 40 with a grey-card or ambient mode, when the camera
    says ISO 100, without blowing RAW highlights.


    --
     
    JPS, Mar 29, 2005
  4. Alan Browne

    JPS Guest

    In message <d29fj0$nlg$>,
    You're seeing the expected *OUTPUT* of a RAW conversion, not the
    *INPUT*.
    --
     
    JPS, Mar 29, 2005
  5. But even with all the parameters set to zero, the Adobe RAW converter is
    gamma-correcting the data and converting from 12 to 8 bits (or the
    equivalent, if you stay in 16 bit mode). If you assume a white
    reference of 2048 (to give one stop of headroom), ARC is calculating
    something like:

    out_val = 255 * (in_val / 2048) ^ 0.45

    If your output value is 118, we can guess that the 12-bit input value
    was probably around 375 (out of 4095) but we can't really be sure.
    (This is just the gamma correction step; ignoring any colour space
    issues).
    Unless you extracted raw pixel data from the file without using ARC, you
    are seeing *processed* or *converted* values. A different raw converter
    would likely give you different output from the same input.

    Dave
     
    Dave Martindale, Mar 29, 2005
  6. It's standard for just about any process that produces a JPEG file,
    because people implicitly expect JPEGs to be gamma-corrected. If it's
    in sRGB, the nonlinear conversion step is explicitly spelled out, but in
    fact sRGB simply writes down what everyone was already doing in this
    respect.

    Everybody is accustomed to working with gamma-corrected data,
    particularly in 8-bit. Even if you convert to 16-bit in Photoshop, you
    still usually work on nonlinear data, because linear data is unviewable
    unless you set the hardware lookup table to do the gamma correction, or
    unless your editor does it for display.

    The Canon raw converter has 3 output formats you can choose: 8 bit and
    16 bit (both implicitly gamma corrected) plus 16-bit linear. The only
    time you'd use the latter is if you wanted to do your own gamma
    correction, or perhaps to do calculations on pixel values.
    I haven't, but I have written software to drive several CCD cameras
    built into film scanners. The mapping from linear to gamma-corrected
    data has to happen *somewhere* before you drop to 8 bits.
    Because nobody wants linear data for pictorial applications.
    No. Not by default, anyway.

    Dave
     
    Dave Martindale, Mar 29, 2005
  7. Yes. If you assume that 100% reflectance white in a correctly exposed
    scene should be at 255, then it follows that 18% reflectance grey should
    be at 255 * 0.18^0.45 = 118. The 0.45 is the standard value of gamma
    correction exponent that's almost universally used (unless you're on a
    Mac). Given this assumption, 118 is the "perfect" grey value.

    However, it's also reasonable to assume a grey of 12% or 16% instead of
    18%. It's also reasonable to make 255 90% reflectance instead of 100%,
    or to put 100% at (say) 235 to provide some headroom. All of these sets
    of assumptions put "mid grey" somewhere close to, but not exactly, 118.

    Dave
     
    Dave Martindale, Mar 29, 2005
  8. You'd be wrong. Virtually every JPEG file ever created has built-in
    gamma correction (call it gamma encoding or power-law encoding if you
    like). Photoshop normally reads and writes this data without touching
    it (unless colour management is enabled) when the source is a JPEG, but
    when the source is camera raw data it has to do the encoding in the
    conversion process to produce a viewable image.

    The Canon raw converter even gives the user the choice of regular (gamma
    corrected) 16-bit data and really linear 16-bit data. If you have
    access to one of the better Canon cameras (G series, Sxx series, or
    DSLR) you might find it interesting working with 16-bit linear. You
    can't even see what's in the image unless you provide lots of gamma
    correction in your graphics card, or immediately do the equivalent of
    gamma correction using Levels.

    Dave
     
    Dave Martindale, Mar 29, 2005
  9. Alan Browne

    Owamanga Guest

    118/4095 <> 118/255

    So I don't know what your point is.
     
    Owamanga, Mar 29, 2005
  10. Alan Browne

    Owamanga Guest

    Nobody is talking JPEG, I don't know why this keeps coming up. The
    conversion is RAW DATA TO PHOTOSHOP PSD FILE. No JPEG!

    If I change my display brightness and contrast using the little
    buttons on the screen, nothing changes to the RGB values in the file,
    or in the info tool when loaded into photoshop.

    If I change my DISPLAY gamma, nothing changes to the RGB values in the
    file or photoshop. If you have a windows machine you can do this test.
    Run Control Panel, Adobe Gamma, grab the slider and move it either
    side of it's current location. Watch contents of all RAW files on your
    system, or even a file loaded into photoshop and the gray area that
    was 118,118,118 remains... guess... 118,118,118. That's because this
    is a DISPLAY gamma, performed at the OS level.
    Having read the sRGB standard, yes, gamma correction (or a non-linear
    mapping) is being performed by the RAW converter to bring it into your
    system's working color space. (Again, this is separate from your
    'display' color space). Interestingly, it suggests that a gray scale
    image doesn't go through this conversion.

    But all along, the 118 number was quoted with sRGB in mind, a well
    documented standard color space with known gamma.
     
    Owamanga, Mar 29, 2005
  11. Alan Browne

    Alan Browne Guest

    The .461 coefficient comes directly from the cipa equations and that
    leads to the 118 value. Within the CIPA equations is a gamma factor.
    Otherwise I believe midtone grey would be 127,127,127.

    There is great latitude for me to be wrong about that of course.

    Cheers,
    Alan
     
    Alan Browne, Mar 29, 2005
  12. Alan Browne

    Alan Browne Guest

    It's not as unbounded as you suggest. The 18% grey is quite valid and
    easy to fix in a scene latitude. Camera meters are anchored by it.
    It's the extremes that are more mysterious.
    I don't agree with the "roughly equate".
    True enough. Incident meters are closer to 12 or 14%. (Recent
    discovery, always thought they were 18%).
    I don't quite agree. It's a valid (or rather 'useful') mid point to set
    exposure. IAC as soon as one angles the grey card with respect to the
    lens axis and the light source, the "18%" itself is not meaningful
    anymore. From there it's about controlling the latitude and
    understanding how a 'brighter' or 'dimmer' looking grey card will affect
    the exposure. From there it's just experience.
    0.461 * 4096 = 1881. No correlation at all...
    The only thing the test at hand needs to illustrate is that there are
    differences between the various cameras. One would expect that all
    cameras from all manufacturers expose the same scene the same way for
    the same exposure at a given ISO. The C d'I test shows they do not. My
    test shows a measured outcome within a couple levels predicted outcome
    (for R,G) and within 10 levels for B. (In the C d'I test the Minolta
    was the only camera consistently "on ISO". I don't have the same
    camera, but Minolta have long been highly consistent in their exposure
    systems so I expect my camera has the same.)
     
    Alan Browne, Mar 29, 2005
  13. Alan Browne

    Alan Browne Guest

    (We been there earlier in the thread. The link I posted (elsewhere in
    thread) gets to the same equation).
     
    Alan Browne, Mar 29, 2005
  14. Alan Browne

    Alan Browne Guest

    This is where, as stated elsewhere, the 118 (rather than 128) comes out
    for mid-tone grey. I agree (now) that the Adobe viewer gamma's the data
    that is presented.
     
    Alan Browne, Mar 29, 2005
  15. Alan Browne

    Alan Browne Guest

    They did not state it properly. They stated examination of JPG images
    which immediately introduces all kinds of questions about all kinds of
    unknowns. I suspect they did the right thing, but explained it poorly
    in the article, perhaps trying to make the difficult simple. I look
    forward to the letters section next month as I'm sure many people will
    be challenging the text. I repeated the test with a known color temp
    source and constrained the data to the least converted path possible and
    got results that were very close to spec with a camera that (per the C
    d'I) is right on the spec. (Caveat: I tested with a 7D, C d'I tested
    with an A200... but Minolta are quite consistent on exposure).

    Per CIPA http://www.cipa.jp/english/hyoujunka/kikaku/pdf/DC-004_EN.pdf
    the 118/256 grey point is clearly defined. All the manufs are members
    of that organization and all the manufs worked on the standard. For
    whatever implementation reasons they don't seem to conform perfectly per
    the the C d'I tests.

    If one measures the height of a dozen people with a measuring tape that
    is off by 10%, you can still tell who is taller or shorter. That is
    really the extent of the validity of the C d'I test as stated. It may
    even be better, but their test description raises doubts.
     
    Alan Browne, Mar 29, 2005
  16. Alan Browne

    Alan Browne Guest

    The C d'I test (and mine) were looking solely at the mid grey point.
    Per the (earlier cited) CIPA doc, one would expect all the cameras to
    come out with the same value.

    If one incident meters (available or flash) and then transfers the
    readings to a camera, then one should expect (according to the C d'I
    test) that the results from one camera model will be different than the
    other. This is the point.
     
    Alan Browne, Mar 29, 2005
  17. Alan Browne

    Alan Browne Guest


    Unity change. All parameters were 'zeroed'.
     
    Alan Browne, Mar 29, 2005
  18. Alan Browne

    Alan Browne Guest

    We're talking on a scale of 0..255 which is what the info in the Adobe
    RAW converter displays. Assuming gamma is included in that info, and
    per the CIPA doc, the 118 (instead of 128) is a result of the gamma
    coefficients.
     
    Alan Browne, Mar 29, 2005
  19. Alan Browne

    Alan Browne Guest

    The point of the statement above is that whatever it displays, it is
    cvertainly not a JPG. The Adobe RAW converter displays the pixel info
    on a 0..255 scale (depite being in 16 bit mode). As all of the RAW
    conversion parameters are set to 0, there is no color offset (except as
    provided for source light color temp).

    I didn't mean to imply that the RAW image was being viewed, only that
    the least amount of change possible had occured to the image data.

    So cool the all caps.
     
    Alan Browne, Mar 29, 2005
  20. If gamma correction was not done, and the data was converted to an 8-bit
    linear space, 18% grey would be at about 0.18*255 = 46.

    The problem is in the meaning of midtone. Our eyes see brightness roughly
    logarithmically, with each doubling of brightness seeming to be about
    the same size change, and with mid-grey about half way along a scene
    brightness range of 5 or 6 stops. A CCD produces half its peak output
    voltage at exactly 1 stop down from max white, a tone that our eye still
    considers part of the highlights of the scene.

    The "gamma correction" step maps the linear output of the A/D converter
    in the camera into a nonlinear space that better matches the way our
    eyes see. With only 8 bits per colour channel, this is important to
    avoid quantization artifacts. If we work with 16 bits/colour, we can
    get away with using a linear representation if we want, but not at 8
    bits.

    Dave
     
    Dave Martindale, Mar 29, 2005
    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.