Light fall off on dSLRs - an experiment

Discussion in 'Digital SLR' started by Kennedy McEwen, Mar 16, 2006.

  1. That is precisely why the test was reduced to *ONE* variable -
    specifically the one variable that you and others claimed was causing
    the effect.
    You really are making things difficult for yourself. You have been
    talking about how a lens functions - how exactly do you think a lens
    design is undertaken without considering how the lens functions? Each
    possible ray path must be traced through the lens.
    I did not say that the light falling on a single pixel could be modelled
    as a single ray, it is clearly a number of rays extending over a range
    of incident angles. I tested the sensitivity of the sensor at two
    extremes of that range and found essentially no difference. The reason
    for choosing the extremes was simple - if there is a difference it is
    most likely to show up at the extremes. It didn't.

    It is just possible, but unlikely, that some intermediate angles do
    produce some significant variation and I certainly plan to test other
    angles in due course. However when, as I now expect, each of those
    lesser incident angles show no essential difference then clearly the sum
    of any subset of all of incident angles will be the same, and thus
    demonstrate fully that your claimed sensor response is completely bogus.
     
    Kennedy McEwen, Mar 17, 2006
    #41
    1. Advertisements

  2. This is a good point. The foreshortening should reduce the number of
    photons per pixel, independent of reflection or absorption
    by the pixel, and independent of Lambertian or not. Think
    of a pixel as a little square hole. As you tilt it, the
    number of photons going through the hole gets smaller,
    because the projected area gets smaller.

    Roger
     
    Roger N. Clark (change username to rnclark), Mar 17, 2006
    #42
    1. Advertisements

  3. Kennedy McEwen

    Battleax Guest

    But he backed up his argument with words like idiot, nutcases, ignoramuses.
    Surely this holds some sway over your intelligent, well researched response
    :)
     
    Battleax, Mar 17, 2006
    #43
  4. Lastly (yes, I'm like a dog with a bone...)...

    I just checked specifications of an 8Mp Kodak CCD sensor, 4/3 size.
    OK, so it's not a great sensor, and maybe it is completely incomparable
    to a Canon CMOS.. (does anyone have a link to the *full* tech specs of
    any Canon or Sony sensor?)

    In those specs, a simple graph shows the sensitivity to light incident
    angle. At just 15 degrees (ie 75 degrees from the sensor face), the
    light sensitivity is down by about 20%. At 25 degrees, it is nearer to
    50%. The graph does not go any further... In fact there is a related
    recommendation of keeping the incident angle to within 12 degrees of
    straight on, to avoid noticable vignetting.

    Hmmm. Who to believe... So is the Canon sensor performance strikingly
    different to this? Does anyone have links to similar specs for other
    sensors?

    The link is:

    http://www.kodak.com/global/plugins/acrobat/en/digital/ccd/products/fullframe/KAF-8300CELongSpec.pdf

    But be warned, it says it is 836 Mb!! Something tells me that must be
    a misprint, but it is clearly a very large file, and I aborted before
    it finished. The graph is Figure 5, I think..
     
    mark.thomas.7, Mar 17, 2006
    #44
  5. It was in there by inference rather than specifically stated. The Canon
    lens mount is 55mm diameter and 44mm from the focal plane, and the
    camera was rotated so that the lens mount cast a shadow in the centre of
    the frame. That gives a maximum angle of around 31.5deg.
     
    Kennedy McEwen, Mar 17, 2006
    #45
  6. Kennedy McEwen

    Jeremy Nixon Guest

    I have absolutely no idea.
    No. I'm simply looking for a test that doesn't introduce changes whose
    effects are not satisfactorily explained.
    No, it's not. I suggest testing film vs. digital by making film vs. digital
    the only thing changed in the test. You can eliminate the falloff from the
    lens by measuring the difference between the two.

    I don't *care* which way the test turns out; I have no stake in it. But I
    would like to know one way or the other.

    Well, that's not even entirely true. Since I think we *are* headed in the
    direction of having 35mm sensors in digital cameras, I would actually prefer
    that you're right about this. (Of course, there are other downsides to the
    35mm sensor format that are put forth, but this is one of them and I'd like
    it to be entirely fictional, but I haven't been convinced that it is.)
     
    Jeremy Nixon, Mar 17, 2006
    #46
  7. Kennedy McEwen

    Jeremy Nixon Guest

    You must really hate being taken seriously.
    Okay. So why not do it? We *can* measure the falloff from the lens as
    opposed to the sensor by simply using film rather than digital.
     
    Jeremy Nixon, Mar 17, 2006
    #47
  8. I am sure you know about Kirchoff's Laws, relating to this
    Incident = reflected + transmitted + absorbed.

    In this case, transmitted is zero, since the sensor is opaque. You have
    accepted that reflected is the same cosine law as lambertian, so
    absorbed, and thus detected, is simply the difference.
    However, few surfaces are perfectly lambertian right out to extreme
    angles of incidence and I doubt that the sensor is either, so your
    reduction doesn't apply.
    Yes - unchanged and the camera was in manual mode.
    That would mean it would be impossible to underexpose a jpeg even in
    manual mode, which clearly is perfectly possible as other posts have
    demonstrated.
     
    Kennedy McEwen, Mar 17, 2006
    #48
  9. Kennedy McEwen

    eawckyegcy Guest

    You'll have to excuse me if I do not understand why a previously sane
    entity is in the process of going into the deep end of irrationality.
    But at least you now appear to be backing away from the abyss...
    For the reasons everyone is telling you: we can only estimate the
    effect, we don't know (but probably can assume) what the
    QE(angle-of-incidence) is for film, and a host of other problems ...
    while on the other hand we now have a perfectly good, trivial, and
    robust way of directly measuring the function for a pixel. Someone has
    handed you a means to obtain easy knowledge on a silver platter. What
    is there to dislike?
     
    eawckyegcy, Mar 17, 2006
    #49
  10. No, a point source at a long distance has a low *variation* in range of
    angles incidence across the field. The average angle of incidence can
    be varied over the larger range that the lens mount will permit to enter
    the camera. As other contributors, both negative and positive, have
    noted, the total range can be varied by +/-31.5deg, which is more that
    any practical lens can achieve in the same mount, with a variation at
    any specific angle of +/-0.035deg across the field.
    With the light source right in front of the mirror then there will be a
    range of angles across the field - however there will also be a range of
    distances from the light source and we know that the light intensity has
    an inverse square law. In addition, the light distribution from the LED
    has an angular variation, which would be added to any sensor angular
    response and inverse square law intensity variation. It would then be
    very difficult, if not impossible, to separate all of the contributing
    variables. By placing the point source a significant distance from the
    focal plane all of the pixels get the same light intensity incident at
    the same angle. That angle can be changed by rotating the focal plane
    relative to line between it and the light source
     
    Kennedy McEwen, Mar 17, 2006
    #50
  11. As Roger has explained, that is not exactly as simple as you think to
    make quantitative measurements. The qualitative observations that
    people who have used the same lens on both film and digital, including
    myself, suggests that any difference which does exist is very small. The
    test was an attempt to measure the only variable that could account for
    such a difference if it was present - sensor angular response. It
    didn't find any in the most extreme case.

    The test is very simple to conduct and it is simple to extend to
    numerous angles, which is why I provided details of it - so that others
    can, if they wish, make the same or similar tests on their cameras. Not
    all digital sensors are the same, but I doubt that dSLR sensors will
    vary significantly. However, for all we know, the small pixel APS and
    4/3 sensors might actually have more of an angular sensitivity variation
    than the larger FF sensors, or CCDs could be worse than CMOS - now that
    would be a turn up for the books! Its a simple test, it measures
    exactly what it claims to measure - angular response of the pixels.
     
    Kennedy McEwen, Mar 17, 2006
    #51
  12. Kennedy McEwen

    Jeremy Nixon Guest

    We can know simply by measuring it, which has the additional side effect
    of being *exactly* the thing we're trying to measure. If you want to
    know the difference between using film and using digital, why not measure
    the difference between using film and using digital, rather than measuring
    something else entirely?
     
    Jeremy Nixon, Mar 17, 2006
    #52
  13. How do you think the lens can create a larger angle of incidence than
    what is achievable through the lens mount? There is nothing to bend the
    light to a greater angle between the rear element of the lens, which
    must be far enough forward to clear the mirror, and the focal plane. The
    rear lens element must fit into the lens mount and thus must be smaller
    than it.

    If you can suggest any way of a lens creating a larger angle of
    incidence on a dSLR then propose it.
     
    Kennedy McEwen, Mar 17, 2006
    #53
  14. Not quantitatively we can't - there are a lot of approximations involved
    which would no doubt be exploited by some as masking any difference. For
    example, even neglecting the response curve of the film, how can you be
    sure that the lens on your film camera stops down to exactly the same
    f/# as on the digital camera in the time between the mirror rising and
    the shutter opening? It might be the same lens, but the actual iris is
    a function of lots of other mechanical and electrical linkages that can
    change when you switch bodies. How do you know the shutter in both
    cameras has the same unevenness across the frame - all shutters vary
    some amount as they cross the frame. As others have tried to, comparing
    film and digital in a fair and meaningful way is not as trivial as you
    seem to think. That is why eliminating as many variables as possible,
    whilst retaining quantitative measurements is a much preferable
    approach.
     
    Kennedy McEwen, Mar 17, 2006
    #54
  15. Kennedy McEwen

    Paul Furman Guest

    Even without pixel depth the angle straight on for half the sensor would
    be a 17mm wide swath of light vs about a 16mm wide swath at 30 degrees
    or roughly 5% loss. I just sketched it out with autocad & counted the
    number of 1mm lines that hit... somewhere in that ballpark. Presumably
    lens designers are already trying to compensate for that and they have
    to let it slide sometimes as part of the give & take of lens design. But
    anyways the LED test should be showing around 5% just due to angle so
    something isn't working. Maybe with the microprisms poking up it doesn't
    act like it's angled light as long as each microlens isn't shading the
    next one?
     
    Paul Furman, Mar 17, 2006
    #55
  16. Kennedy McEwen

    eawckyegcy Guest

    The whole point of this exercise has been to question the theory that
    there is something special about digital sensors re: high angle of
    incidence, vignetting, etc.

    An experiment was proposed, results published. If the results stand,
    no further measurements are needed: any observed vignetting must be
    from the lens, not the sensor. Using a different protocol is possible,
    but at far higher cost comparitively speaking. Why do more work for
    the same result?

    Now if you want to know how film performs, well, that's something you
    can undertake I guess.
     
    eawckyegcy, Mar 17, 2006
    #56
  17. Kennedy McEwen

    w.beckley Guest

    This is true, and this could even be the difference that he measured.
    In such a case though, would this exposure difference not be identical
    in film?

    Will
     
    w.beckley, Mar 17, 2006
    #57
  18. Kennedy McEwen

    w.beckley Guest

    This is true, and this could even be the difference that he measured.
    In such a case though, would this exposure difference not be identical
    in film?

    Will
     
    w.beckley, Mar 17, 2006
    #58
  19. Kennedy McEwen

    Paul Furman Guest

    Another thing that might be a factor is the LED light is not focused & a
    blurry image is going to spread out... although holding the light back
    may fix that, I don't know. Light falloff could be obscured through a
    lens by throwing the focus way off because that spreads things around.
     
    Paul Furman, Mar 17, 2006
    #59
  20. Kennedy McEwen

    Jeremy Nixon Guest

    The problem is: you're saying that the people who design and make the
    sensors themselves are either wrong or lying about this. Perhaps they
    are, but this test has not convinced me that is the case.
    You're also suggesting that the sensor designers never thought of this
    simple test, and simply agreed among themselves that the angle of
    incidence is important without actually verifying it.

    I'm sorry, but I'm going to need a little more to believe that.
     
    Jeremy Nixon, Mar 17, 2006
    #60
    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.