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

    Alan Browne Guest

    I quit. And not becasue you're right.
     
    Alan Browne, Mar 28, 2005
    #81
    1. Advertisements

  2. C'mon Allan. Stop this before you go too far.

    You might not believe what I say. And you might
    not even like my posts. But please stay civil.


    /Roland
     
    Roland Karlsson, Mar 28, 2005
    #82
    1. Advertisements

  3. Alan Browne

    Alan Browne Guest

    Too far for what? Will you plonk me?
    The purpose of the OP was related to camera ISO settings being correct
    wrt to exposure. That's all. You've taken it well outside that
    definition. Looking at the 18% point is the only comparative means of
    checking (for reasons stated) across sensor types. Beyond there one
    could look at which sensor goes further into the shaddows or higher into
    the highlights but that is not the purpose of the article. I do agree
    that the article mention of JPG is contentious, but doing the tests the
    way I did them eliminates all variables other than color temp
    (corrected) and gamma.

    In my reply to Dave Martindale I stated that the gamma is of no
    consequence, but on reflection it is, I believe, the reason why the 18%
    grey is at 118 rather than 256/2 (128).
     
    Alan Browne, Mar 28, 2005
    #83
  4. For some reason you believe Dave Marintdale when
    he says the same thing.

    IMHO - You have misunderstood how color spaces and mapping
    from RAW to a certain color space works. And when I try to
    explain it you just have to defend yourself instead of
    listening. Bad choice IMHO. But that is how it is sometimes.

    See you in some other thread, with a new blank opinion
    about each other I hope.

    See ya,
    Roland
     
    Roland Karlsson, Mar 28, 2005
    #84
  5. Alan Browne

    Owamanga Guest

    Can you cite the non-linear aspect of this particular conversion the
    Adobe RAW importer is doing?

    Have you browsed the source code? (it is available on the net), and
    seen the part of code that does the non-linear magic?

    If no curves are used on import as Alan has said, and he made no other
    exposure adjustments, why wouldn't the conversion be *roughly* linear?

    Isn't that what we want and expect from a RAW importer?
     
    Owamanga, Mar 28, 2005
    #85
  6. Alan Browne

    Owamanga Guest

    I'd suggest again that gamma correction is done for display purposes
    only and won't affect the reading the info tool has for a particular
    pixel. Eg, photoshop doesn't re-map the image based on gamma
    correction, Adobe does this at the OS level instead.
    A big IF.
    Agreed. IF.
     
    Owamanga, Mar 28, 2005
    #86
  7. Alan Browne

    Alan Browne Guest

    Let me change the above to:

    I stated that the gamma is of no consequence, but on reflection and I
    believe confirmed by the below, is the reason why the 18% grey is at 118
    rather than 256/2 (128).

    http://www.cipa.jp/english/hyoujunka/kikaku/pdf/DC-004_EN.pdf
    refers on page 6.

    Cheers,
    Alan
     
    Alan Browne, Mar 28, 2005
    #87
  8. Alan Browne

    Owamanga Guest

    But, if I display those two images side by side, one containing 132
    and one containing 118, the 132 image will look overexposed compared
    to the 118 image. Thus, an EV correction would be needed on the 132
    equipment to bring it back in line.

    This is the whole point of known ISO values.
    Due to the complexity, and Alan's constant disclaimer regarding
    DMax-DMin (or shoulder/toe) he isn't attempting to cover all zones,
    just one: The gray card reading.
     
    Owamanga, Mar 28, 2005
    #88
  9. That was not what I meant. I think it would be a shame if
    the arguing is going out of hands. It is not _that_ important.
    We will meet in other threads. It would be a shame
    if we had biased opinions on each others post.
    This is one of the main points in my postings.
    You have to take the unlinerities and also
    the biases in the conversion when evaluating the
    number 118. The number 133 might be just as
    good. It depends on the mapping.

    Now - I undertand what you try to accomplish - and I appreciate
    that. And - under some limited circumstances - it works. But
    - there are weeknesses IMHO. I tried to point those out.


    /Roland
     
    Roland Karlsson, Mar 28, 2005
    #89
  10. Alan Browne

    Owamanga Guest

    That's the sRGB gamma characteristics, not those of your display. I
    still maintain the brightness, contrast (you can even turn the damn
    screen off) or system gamma settings will make a jolt of difference to
    the pixel RGB value that the info tool shows you in photoshop.

    You (I think) already clarified that your tests were in sRGB color
    space with gamma characteristics outlined in IEC61966-2-1.
     
    Owamanga, Mar 28, 2005
    #90
  11. Message-ID: <[email protected]>
    User-Agent: Xnews/5.04.25

    I expect a linear import from the RAW file when converting from
    Bayer to plain linear RGB. It is not 100% sure - but it probably
    is very near to linear.

    The unlinear aspect is only with regard to the 118/255 value.
    I assume that this works the same as in Photoshop. Adobe RAW
    importer probably uses some color space (e.g. sRGB or Adobe
    RAW) for its 8 bit representation. In Photoshop this color
    space is called working color space. You can have another saving
    and a third display and a fourth printing color space. But
    the numbers shown when hovering over pixels in Photoshop is
    according to the working color space.

    Now - the color space is almost always a non linear color
    space - i.e. it has some gamma mapping or some other means
    of compressing information into only 8 bits.

    The reason that I am very sure that Adobe RAW converter does
    the same as Photoshop is that the resulting JPEG file also
    has 118 in the same area. And the JPEG probably is Adobe RGB
    or sRGB and those have non linear mapping.


    /Roland
     
    Roland Karlsson, Mar 28, 2005
    #91
  12. Alan Browne

    Jeremy Nixon Guest

    Well, what do you get, in photographer-language, if you cut a light to 18%?
    You get 2.5 stops less light. So, assuming my above simplified scenario,
    your 18% grey would be halfway between 1024 and 512, or 768.

    That you clearly want (and get) more than two and a half stops above middle
    grey, though, is the first clue that there is something else going on here.

    See, the thing is, 18% grey doesn't mean a whole hell of a lot in terms of
    photography. Oh, I know everyone swears by that Holy Number, but that's
    just because everyone else does. The 18% grey card has become a piece of
    cargo cult knowledge, repeated so often that everyone takes it as a given.

    "18% grey" is a measurement of reflectance. Your light meter cares about
    luminance. They're not the same thing. You can, however, roughly equate
    them and come up with some reflectance number that sort of equates to a
    certain luminance value, which is why it all can work out.

    But is your light meter calibrated to 18% grey? Probably not. That is,
    after all, not what standard light meters are calibrated to, despite the
    fact that just about everyone will tell you that they are. 18% probably
    comes from print, where I guess that really is somewhere close to the
    middle point, or something.

    Ansel Adams argued for 18% grey in photography. He wanted light meters
    to be calibrated there. He was not successful. The standard for light
    meters is in fact roughly 12% grey -- it is specified in luminance, not
    reflectance, but that's about where it translates. That's what your
    light meter is telling you. Well, probably not, exactly; different
    companies do things differently. Nikon meters at 12% grey. Others may
    vary. There is no way, however, that 18% is anywhere near correct.

    So, first, we need to take this old idea of "18% grey" and toss it into
    the trash along with all the other bad photography advice that's been
    thrown around since the invention of the art.

    At that point, your middle point of 12.5% (to make it mathematically
    symmetrical) ends up falling at 512 on my simplified scale.

    But then, do you really want it there? Clearly, once you are recording
    linear light values, there are myriad ways to translate them into a
    picture, and "correct" math won't mean much to someone who just wants
    his pictures to look nice. So, you might end up with your midpoint
    somewhere else entirely, depending on what the camera engineers settled
    on for the best results. And, of course, they don't tell you what that
    is, because they obviously think their solution is the best one and they
    don't want the other guys getting any ideas.

    Anyway, what's the bottom line? I think it's that the "midpoint" is
    pretty elusive, and any attempts at controlled testing need to take
    that into account.
     
    Jeremy Nixon, Mar 28, 2005
    #92
  13. It is here that we disagree. There is much more to an ISO
    value than this. The sensor has a certain exposure latitude
    and the scene you want to record has a certain range. At
    the uppermost part of the exposure latitude you get
    saturation and at the lower parts you get noise, both
    analog noise and digital noise. The best exposure is
    an exposure where the important parts are found where
    the noise is low and also far away from clipping.

    If the scene is flat - then there are a whole set
    of exposures that is good - as the solid state sensor
    is very linear. If the scene is high contrast - then
    it is very hard to find the best exposure - you might
    lose quality at both ends - it is a compromise. You
    really would like to decrease the contrast of the sensor
    - but that is not possible. With film you could choose
    a softer development.

    As can be seen in the Kodak article - there are three
    different views on ISO; noise limited, saturation limited
    and the ISO that is used by the exposure meter.

    For negative film - the choosen ISO is noise limited - for
    positive film the choosen ISO is saturation limited. For
    digital we can choose either - as it is so linear. It
    is up to the manufacturer of the camera to choose method.
    Yupp - and I appreciate that.

    But knowing about the zone system - and also have been using it -
    I know that the knowledge about just one zone in the middle is not
    all that useful.

    I have tried to find the best method to make good pictures with
    a digital camera and I have found that any picture, regardless
    of how correct the exposure, needs to be adjusted with levels.
    It is exactly the same as changing the development time or/and
    paper grade when applying the zone system.

    All this said - it might be a good rule (if you use sRGB) that
    a grey card found in a normally lit part in the picture
    shall have the value 118 _after_ doing the neccessary manipulation.
    I have not tested that. It might be so.


    /Roland
     
    Roland Karlsson, Mar 28, 2005
    #93
  14. Thx Jeremy!

    I could not have put it better myself.


    /Roland
     
    Roland Karlsson, Mar 28, 2005
    #94
  15. Alan Browne

    eawckyegcy Guest

    If the linear->nonlinear transfer function is sRGB. See:

    http://www.w3.org/Graphics/Color/sRGB

    Simple evaluation of the composition of equation 1.2b and 1.4 in the
    above URL for a linear value of 0.18 returns 255*(1.055*(0.18**(1/2.4))
    - 0.055) = 117.65 ~= 118.
     
    eawckyegcy, Mar 28, 2005
    #95
  16. Alan Browne

    Owamanga Guest

    Okay, but unless we can control the contrast in the scene (and often
    we can't), we choose an exposure based around an 18% gray card
    reading, nudging either way as required to keep at least the 'subject'
    part of the scene from banging in to either end of the histogram.
    Should have read this all first, You just said the same thing, but
    better.
    Okay, but that's what the digital camera gray cards are for (the ones
    with black, white and gray). Still not every zone, but more helpful
    than just zone V.
    I completely agree. But photographers (like Alan) are striving for
    some kind of standard by which they can fit in a new piece of
    equipment into their many years of experience, in particular the
    'correct' exposure for a given situation. There must be at least a
    guideline he can develop that identifies the differences between the
    way the two mediums behave, and therefore what 'rule-of-thumb'
    adjustments may need to be made when moving between them.
    I think the point is to try and get it there _before_ doing any
    manipulation. Any systematic manipulation you find yourself doing to
    the images (for example, on a Nikon D70, I tend to need a +0.5 to +1.0
    EV adjustment when importing RAWs) indicates a systematic error in
    exposure.

    Further adjustments either way based on the Zone system, and what
    effect you want should then be applied without worry to 118/255 or
    such measurements. This is the 'artistic' bit of the exposure choice.
    Not having a huge set of grey-card RAWs at hand, I can't do this test.
     
    Owamanga, Mar 28, 2005
    #96
  17. Alan Browne

    eawckyegcy Guest

    .... only if the linear->nonlinear transfer functions in all the cameras
    are identical. Did they actually prove this?
     
    eawckyegcy, Mar 28, 2005
    #97
  18. Alan Browne

    Owamanga Guest

    Not so much in the camera, but in the RAW converter. If we all use
    Adobe RAW importer and we ask it to convert to sRGB color space, then
    I would guess the functions would at least be similar, if not the
    same.

    Other importers do (I know from playing) some fairly funky stuff.
     
    Owamanga, Mar 28, 2005
    #98
  19. But - in general - when the lighting is not constant - this
    is rather hard to do. And - I guess that the histogram method
    will become more and more used - thereby the ISO setting becoming
    less and less important. Its easier - and leads to the best result.
    And if you use automatic matrix metering - then you have not
    the slightest idea what the camera does for mumbo jumbo - so the
    ISO is also there not really known.

    Only for external meters, strobes and flashes is the actual
    ISO really important to know.
    This would hint at an under exposure. It could also hint at the
    camera not wanting to saturate the sensor and your converting
    program assuming an exposure aimed at a near to saturated sensor.
    This might be a good choice made by the camera if the noise
    of the sensor is low. Parts of the pictures that hits the
    saturation level cannot be saved.


    /Roland
     
    Roland Karlsson, Mar 28, 2005
    #99
  20. Alan Browne

    JPS Guest

    In message <d27daj$oqj$>,
    How were you viewing the RAW data? RAW CONVERTERS DO NOT SHOW ANY RAW
    DATA! Not ACR, not C1, not RSE, not PSP9, not PSE 3.
    --
     
    JPS, 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.