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

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

  1. The same nonlinear gamma correction is needed *somewhere* along the way
    to get a viewable output file, whether it's PSD or JPEG or TIFF or PNG.
    So the raw converter does it automatically for you.
    That's irrelevant. This "gamma correction" isn't actually correcting
    for anything in the display. It is a fixed nonlinear transformation
    needed to make reasonable use of the 8-bit intensity representation
    used. I prefer to call it "gamma encoding", to emphasize that it isn't
    correcting for anything at all, but that's not the common terminology.

    Even if you have an inherently linear output device (e.g. DLP
    projector), you *still* want to use gamma encoding in the data file.
    It's just a happy accident that the power function which approximately
    corrects for CRT electron gun non-linearity *also* happens to be about
    the best available transfer function for storing pictures in 8 bits per
    No one has suggested it does, because the image file gamma *encoding* is
    NOT a *correction* for anything in the display system. It's not supposed
    to be a correction. However, by being mathematically close to the
    correction that is ultimately needed by a CRT, it minimizes the amount
    of correction power needed in the graphics card hardware.
    Right. This gamma correction, which really *is* correction for a
    specific display, acts along with the gamma *encoding* built into the
    image data, to give the overall function that maps CCD output voltage to
    CRT gun input voltage.
    Where does it suggest that? Greyscale images are normally gamma encoded too.

    Dave Martindale, Mar 29, 2005
    1. Advertisements

  2. The output of a raw converter is far from an identity function, and far
    from a linear transformation, even with all the tweakable parameters set
    to their defaults. The exception: Canon's raw converter when the
    special 16-bit linear output mode is selected.

    Dave Martindale, Mar 29, 2005
    1. Advertisements

  3. Alan Browne

    Alan Browne Guest

    Could be. My assumption (of 127) is based on memory of other
    discussions in the past. No idea if gamma is in those or not. 46/255
    seems very low.
    I hear ya. Just not convinced that 18% grey is at as low a point as
    46/255. Exponentials do have a habit of biting though.
    Alan Browne, Mar 29, 2005
  4. Alan Browne

    Alan Browne Guest

    I don't suppose that's available for free AND reads the Maxxum 7D?

    Alan Browne, Mar 29, 2005
  5. Alan Browne

    Owamanga Guest

    Sorry, wrong document. The CIPA one (I can only find the draft, Alan
    had a link to the final). Top of page 5 mentions target data can be
    from one of two sources:

    1. A monochrome image (no mention of gamma)
    2. Luminance signal calculated using Gamma.

    I can't cut & paste the full text (protected PDF) but it's here:
    Owamanga, Mar 29, 2005
  6. Alan Browne

    andrew29 Guest

    No, that's not how Photoshop works. The gamma of the working data is
    the gamma of the working space, which is 2.2 for both Adobe RGB and
    It's 118/256 ^ 2.2 = 18.2%.

    andrew29, Mar 29, 2005
  7. This is only true for one case - a scene that is evenly lit
    and without specular highlites you want to catch.
    Yes - that seems lika fair assumption. But - it is not so and
    never has been so. You always have to callibrate your camera if
    you want to use external light meters or flashes. Very irritating.

    Personally I think that this very old problem now has a
    rather nice solution - use the histogram and the clipping
    warning in the LCD. No need for callibrating and guessing
    and heuristics and all that jazz

    Roland Karlsson, Mar 29, 2005
  8. JPS has a history of becoming agitated when he knows he is right :)
    So have I - but I have learned to calm down before writing
    several years ago - almost always :)

    JPS is right this time also (he usually is :). The scale 0..255
    shows the value that the RAW data should be mapped to if it was
    converted to an 8 bit file using the current color space - which
    is sRGB I think.

    It is a computed value. It is cumputed exactly the same way that
    the actual 8 bit value will be - before doing the JPEG compression.

    The actual RAW value is something that is approx 18% of 4K.

    Roland Karlsson, Mar 29, 2005
  9. wrote in
    According to the paper - this should be approx the same as

    255 * 0.18**(1/2.2) = 117

    Yepp - seems so.

    This assumes that 100% white is at 1.0 and black at 0.0.

    So - it assumes that you expose a white paper to 90% of
    saturation in the digital domain.

    Roland Karlsson, Mar 29, 2005
  10. Alan Browne

    Jeremy Nixon Guest

    A measurement of reflectance only translates to a particular luminance under
    specific conditions. So it's never going to be exact.
    The allowable variance is 2%, I think, so 12-14% is about right, but 18% is
    of course never right.
    Sure; you're doing a lot of things and using a lot of experience to
    compensate for the fact that the card is the wrong color. First you're
    angling the card toward the light source, so "18%" is no longer true,
    and then going on from there. I, however, would say that knowing how to
    compensate for something being wrong doesn't actually make it correct.
    Depends how you define the ISO, I expect. There's another ISO/ANSI
    standard that says how that's measured and determined, if you want to
    pay for it. I don't, so I haven't read it, but there's no way in hell
    it uses 18% grey as the reference point.
    Jeremy Nixon, Mar 29, 2005
  11. I have not ssen the standard either - but I have seen the abstract.
    It defines three different ISOs: noise based, saturation based and
    "actual choice". I am 100% sure what the two first are - and no one
    is using any grey card - nor 12% neither 18%. The third I can only
    guess - but it is probably based upon the two others.

    Roland Karlsson, Mar 29, 2005
  12. Alan Browne

    JPS Guest

    In message <[email protected]>,
    If only the histograms were more accurate; even if you are shooting
    jpegs, they can easily miss clipping in the red and especially the blue
    channels. On most cameras that have RAW, they only show you the JPEG
    luminance histogram.
    JPS, Mar 30, 2005
  13. Alan Browne

    JPS Guest

    In message <>,
    My point is this:

    I went to sleep early, "woke up" at 2AM, and went to the computer before
    I fully woke up!
    JPS, Mar 30, 2005
  14. Alan Browne

    JPS Guest

    In message <d2bu1q$3vu$>,
    I think that even that is not the RAW data, scaled. The blackpoint is
    already subtracted, and color-balancing is applied, IIRC.
    JPS, Mar 30, 2005
  15. Alan Browne

    JPS Guest

    In message <d2bnin$h8l$>,
    It is in the same class of image - gamma-adjusted, clipped,
    color-balanced data. The JPEG compression is trivia.
    Same standard used with JPEGs.
    Changing anything from those zeros doesn't make the conversion any more
    "changed". You seem to be imagining something that isn't true - there
    is no one "unchanged" RAW conversion. It is always interpretation.
    JPS, Mar 30, 2005
  16. Alan Browne

    JPS Guest

    In message <[email protected]>,
    .... more like blackpoint + 0.18 * 2000 for the green channel, and
    blackpoint + 0.18 * 2000 * f for the red and blue channels, where f is
    the channel's relative capacity to pass white light. At least this is
    roughly how the 10D and 20D work. The red channel is about a stop less
    sensitive than the green on the 10D, so 18% red on the 20D would be only
    about 180 above blackpoint, out of about 3975 levels, or about 4.5%
    luminance. 18% green would be about 360, or about 9%.

    Moral: Don't take under-exposure lightly!
    JPS, Mar 30, 2005
  17. Alan Browne

    Alan Browne Guest

    I know what result I got under controlled lighting, aperture and ISO
    setting. It is too close to the C d'I result to ignore. Their result
    traces back to the CIPA standard. So, either a coincidence or a
    confirmed result. Please feel free to do the same.

    Alan Browne, Mar 30, 2005
  18. Alan Browne

    Jeremy Nixon Guest

    The problem with the test method is that ISO sensitivity has absolutely
    nothing to do with 18% grey anything. So, starting the test from that
    assumption dooms it to be forever questionable. Repeating their test
    is fine, but you're not measuring ISO sensitivity, you're measuring
    something else, something more like "the exposure level at which 18%
    grey comes out the other end the way I expect it to".
    Jeremy Nixon, Mar 30, 2005
  19. I think it's freely downloadable, but only works on Canon raw images.

    But you might look into dcraw, a free/open source raw converter.

    Dave Martindale, Mar 30, 2005
  20. Probably. But in that case, data values in the 16-bit data *are*
    linearly proportional to scene light intensity, something not true when
    any sort of gamma encoding has been done. It's not raw A/D output, but
    it's close to raw light measurmements with an unknown scale factor.

    Dave Martindale, Mar 30, 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.