Light fall off on dSLRs - an experiment

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

  1. Kennedy McEwen

    Paul Furman Guest

    I was just saying if you count the blue & magenta 1mm lines on this
    there are more 'magenta' photons hitting the sensor at 90 degrees than
    blue ones at 20 or 30 degrees off. Does that diagram represent your test
    where half the sensor was lit up at an angle? My dimensions are an
    approximate Nikon mount so maybe a little different like the 2.5%.
    Whatever the exact number the test should show at least that much.
    Microlenses couldn't help that basic photon count. Anyways it is clearly
    a trivial amount.

    But what about that big digital back view camera? Why no microlenses on
    Paul Furman, Mar 17, 2006
    1. Advertisements

  2. G'day, Philip.

    Message ID's? Try mesages 21 and 51 in this thread (sorry, I'm an
    uneducated Google user!)

    Or tinyurls:

    Anyway, essentially I have the following concerns:

    1. I am unconvinced that the angle you can achieve by shining a light
    past the lens mount is as steep as that from the rear element of
    whatever wide angle lenses are allegedly causing the problem. Yes, the
    rear element has to clear the mirror, but it nevertheless sits inside
    the lens mount, and is closer to the sensor. It would take careful
    measurements to determine these angles, but here the lens has been
    taken out of the equation, even though we know the problem is claimed
    to only apply to some lenses (presumably those with very high incident
    angles). I do not follow this logic, and it leaves a hole in the data.
    What angle allegedly causes the problem? If you can't answer that,
    then you are in trouble before you start.

    2. A lens constrains almost all of its light output to the image
    circle/cone. With this experiment, the light is simply falling through
    the lens mount. There could be reflections/light spill *that would not
    exist* if the lens was there. Even Kennedy's own description suggests
    that may be the case - "the focus screen diffuses the lens mount
    shadow quite a lot". Huh? Perhaps he means the *anti-aliasing* screen
    makes it difficult to see, or perhaps the underside of the
    mirror/mirror chamber walls or something else *is* causing light
    spill... He has not used a collimated or restricted light source, so
    it cannot be determined whether all of the light hitting the sensor is
    coming *directly* from the light source. This seems to be a major

    3. Sensor specifications *do* indicate reduced light sensitivities at
    as little as 10 degrees off axis! I'm not claiming that the Kodak 8Mp
    CCD sensor I quoted is necessarily typical, but links and references to
    this issue are out there. Typically for CCD's, light loss of 10-20%
    can be expected at angles around 15 degrees, and 30-45% at 25 degrees!
    I question whether the Canon CMOS sensor is that much different, but I
    don't know - I can't find any Canon CMOS specs online.

    I have listened to the "other problems" that would be introduced by
    instead performing the experiment by using a wide angle lens at exactly
    the same settings on a digital FF DSLR, and then a film SLR. The
    'variables' I have seen quoted so far, like inaccuracies in the
    shutter/aperture from camera to camera (how would that affect the
    result, again?), are of such little significance to be irrelevant. I
    keep coming back to the fact that the whole problem is (allegedly)
    increased visible 'vignetting' when using a wide angle lens on a FF
    DSLR, over that seen on a similar film camera.

    This test does *not* address that primary (image quality) issue. We
    haven't even seen the resulting images (exciting as that would no doubt
    be!). It instead tries to prove that incident light angle is
    irrelevant, and uses a very crude method that IMHO is flawed in a
    number of ways, outlined above. And I don't mean to insult Kennedy,
    who has a good history of posting good information, but he has made
    these statements before and therefore has a bit of an incentive to
    'prove' himself right.

    So I remain unconvinced, and plead with someone.. anyone!.. who has
    both Canon film and FF DSLRs and a suitable wide angle, to try the
    *other* approach. Rigorously and carefully... (O:

    The only (largely irrelevant) comparison I have seen is between a Nikon
    D2X and 1DS MkII, here:
    (scroll down to 'Wide Angle Issues')

    While this shows quite bad vignetting from the Canon, it doesn't
    address the original issue (and looks pretty unscientific too..!)

    Anyway, we should all stop fussing and go take some pictures... (O:
    mark.thomas.7, Mar 17, 2006
    1. Advertisements

  3. Thanks for that post!
    EXACTLY!! Like I said *many* times above - the fact that the light in
    Kennedy's experiment was not collimated and restricted could be *very*
    significant. The lens restricts the light almost completely to the
    image 'cone'. But once you remove that lens, and just throw light in
    to the chamber, all bets are off.. You now have a spread of light
    falling into and onto everything, spilling and reflecting in ways that
    would be very difficult to measure or predict.
    mark.thomas.7, Mar 17, 2006
  4. True enough.
    Okay, that's a pragmatic argument that I'm reasonably happy with.
    And, of course, the limitation was initially set by making the test
    easier. The angle you get is steep enough to be interesting, which is
    enough to justify your first round of testing easily enough.
    David Dyer-Bennet, Mar 17, 2006
  5. Worked for me.

    *Thank you* for supporting your theories with your own tests! This is
    how we *really* learn to understand things.

    Is there some utility that produces the real sensor data from a RAW
    file for your or (I hope) many formats, or did you have to write your

    Yes, I was considering proposing that a time or two. Then you're only
    bothered by reflections off the sensor that bounce around.

    This is all fascinating.
    David Dyer-Bennet, Mar 17, 2006
  6. You probably mean, as others have suggested, that I did not use a
    You are missing the point.

    When a lens is fitted, the light rays are largely constrained to the
    image cone. Nothing much hits the chamber walls, the lens mount, the
    mirror underside, and whatever else is visible when you have no lens.

    With the lens off, there is light going into the chamber in a way it
    was never intended to (by angling the camera you have made it worse.)
    The camera designer will not have paid much attention to reducing light
    spill/reflections for such a situation. You might even have a lovely
    bright reflection from the inner edge of the lens mount, for all we

    The light needs collimation *so it will act like light from a lens, and
    be constrained within the 'normal' limits*.

    This ISN'T about the photons that strike the sensor directly from the
    light source!!! It's about any potential *other* photons that are
    coming from who knows where. Your comments on this issue (as a lens
    designer) would be appreciated. (O:
    mark.thomas.7, Mar 17, 2006
  7. And an opaque sheet with a small hole in it can eliminate potential
    reflections from other things in the room pretty easily, come to that.
    It seems entirely possible that Canon made special efforts to handle
    falloff on the full-frame sensor; there are steeper angles inherently
    involved, *and* it was intended for their top-end pro body where
    people will care the most. I wish someone with a DCS-14n would pop up
    about now....
    Knowing the cause doesn't directly determine what to measure anyway.
    And being sure you know exactly what's going on from the start is
    questionable science, at best.
    David Dyer-Bennet, Mar 17, 2006
  8. You know, I'd never really thought about *how* microlenses function.
    I think I do now; thanks!
    David Dyer-Bennet, Mar 17, 2006
  9. Kennedy McEwen

    JPS Guest

    In message <>,
    If the premise is that light striking the sensor at angles from the
    perpendicular gets captured weaker, then what difference does it make if
    the angular light comes through a lens or not? If there is a
    difference, it is because of the lens; not the sensor. The lens can't
    throw a curve-ball!
    JPS, Mar 17, 2006
  10. Kennedy McEwen

    JPS Guest

    In message <>,
    Well, I wouldn't call it an idiot theory; just an uninformed one.

    Perhaps the contrast of digital images in the range of falloff is higher
    than for film; hence the digital FFs look worse (or expectations were
    JPS, Mar 17, 2006
  11. Kennedy McEwen

    Jeremy Nixon Guest

    Too many variables are changed to make the test at all convincing. I don't
    believe it to be valid.
    I would like for that to be true, but this test has done absolutely nothing
    to convince me.
    Jeremy Nixon, Mar 17, 2006
  12. Kennedy McEwen

    Rich Guest

    I wish people would leave the sensors out of the discussion
    completely. Even if there were some validity to saying a sensor
    "caused" fall-off, its contribution (compared to a lens that CANNOT
    support a decent sized image circle) would be so minor as to be
    Rich, Mar 17, 2006
  13. Kennedy McEwen

    Rich Guest

    Its only relative if you decide to use another lens to magnify the
    image coming from the lens. At "prime" magnifacation (the native
    manification of the lens), the diffraction effects are not visble.
    Rich, Mar 18, 2006
  14. Everyone? I don't hear many people who have experience of both FF
    digital and film saying anything of the sort - they just get on with
    life in the knowledge that there isn't any difference.
    Kennedy McEwen, Mar 18, 2006
  15. Because you don't understand anything about optics.
    Because you don't understand anything about optics.
    Kennedy McEwen, Mar 18, 2006
  16. Kennedy McEwen

    Skip M Guest

    He has removed the largest, most intrusive variable of all, the lens. There
    can be no discussion of how a lens may, or may not, have affected the
    result, because a lens was not used.
    I can't understand why not. If the lens is not part of the equation, and
    the light fall off on the edges of the sensor is minimal, then it must be
    the lens that produces the problem. If he had left a lens, event the best
    lens, on the camera, then the lens used would become part of the discussion.
    BTW, the "I can't understand why not" is not confrontational, I've been
    reading this thread with both interest and trepidation, because there are a
    lot of people here who understand far more about this than I do. But the
    ones that I read that seem to make the most sense are the ones like Kennedy
    and Roger. Your comments don't explain, to me, why you remain unconvinced,
    so I'm really looking for an explanation.
    Skip M, Mar 18, 2006
  17. Kennedy,
    Let's say the light source is to the right of the camera, when viewed
    through the viewfinder. Then at the highest incident angle,
    the right side walls inside the camera are in shadow.
    But what about light entering the camera and illuminating
    the left side inside the camera? That light would
    scatter onto the sensor. You could evaluate this by
    looking at the levels just inside the shadow.

    A better test would use an aperture stop in front of the camera
    that stays perpendicular to the light beam. The spot, say
    0.5 cm diameter on the sensor would record the light, and the area
    around the spot could be used to measure scattered light.
    Subtract off the scattered light. Do all work in 16-bit
    linear conversion.

    Roger N. Clark (change username to rnclark), Mar 18, 2006
  18. If that is what you think then you are badly mistaken! Take a look at
    the mirror box in your SLR - film or digital makes no difference. Do
    you understand *why* it is the shape and finish that it is? It
    certainly isn't because someone like me would take some images without a
    lens fitted!
    If there was then the shadow of the image would not be black. It was -
    buried in the noise floor of the sensor!

    I must be extremely fortunate that all of the scattered light that is
    bouncing around inside the camera just happens to all end up on the area
    of the focal plane that is lit directly.
    No it doesn't. At most it requires elimination of forward scattering
    sources in the path from the LED to the focal plane. This can be
    achieved in many ways - black sheets, as another tester used, nothing
    but free air in the light path, as I did, or a simple card aperture in
    front of the lens mount operating as a field stop. None of these
    improve the collimation of the light.
    I am not a lens designer, and I hope I haven't given you or anyone that
    impression. I design imaging sensors primarily for military systems,
    including visible and thermal imaging cameras. I work very closely with
    people who *are* lens designers - although they would be offended by
    that, since they design much more than optics in lenses!
    Kennedy McEwen, Mar 18, 2006
  19. It could be, but it wasn't. If there had been any indication of light
    being scattered into the shadow area then I would have taken steps to
    reduce it. I just didn't have to because there simply wasn't any of the
    scattering present that you guys are worrying so much about. Zilch!

    I admit, I should have included the shadow levels with the original
    results, but at the time I really didn't think it was important - maybe
    I expected to be better recognised as knowing whether it was a problem
    in my data or not. Nevertheless, there was no significant scattering
    present in this test - and there were far more important things which
    had to be accounted for in conducting the test that nobody even seems to
    have questioned.

    Perhaps, rather than hypothesising about what went wrong with my
    measurements you should try to repeat them yourselves and get some idea
    of what parameters are important to control in such assessments.
    Until it hits the focal plane (or AA filter) and becomes anyone's guess
    where it ends up. If you think you get a nice cone from the lens then
    you are kidding yourself. Cameras are not just smooth boxes for a
    specific reason!
    Rubbish. Try repeating the test and then come back and tell me what
    your biggest problem was - it certainly won't be scattered light if you
    set it up as I explained! Enough of your endless hypothesising about
    what might have happened when none of the results even suggest it could
    have happened that way.
    Kennedy McEwen, Mar 18, 2006
  20. Interesting results and they seem to differ considerably from mine which
    I have just repeated over a range of angles to get more of a response
    map with azimuth and elevation angle of the light source relative to the

    First though, a couple of comments on some of the other points you made.
    I managed to do this by temporarily fitting a lens to the camera body
    and then just aligning it so that the light source was exactly on the
    central focus marker. I then removed the lens for the rest of the
    I wonder if this is just due to a different sensor - it is dramatically
    different from my results which have a much flatter response.
    OK - I used a darkroom with the light source stuck in the middle of a
    wall (one advantage of using a LED - it can be held in place with
    Blu-tac with just the two wires dangling down to the power source). The
    floor of the room is very dark blue, almost black, and whilst the walls
    are off-white they are at least 6-7ft on either side of the light
    source. I did clear things out of the way that could have been
    reflectors, so that cut that problem down a lot.
    No! A laser pointer is a coherent source and you will get large
    variations just due to speckle caused by random points of constructive
    and destructive interference.

    If you want to create a collimated source, use an old film camera with
    the back and shutter open. Cut a small frame size card, with a small
    hole in it and put your light source behind that. Fit as long a focal
    length lens as you have available. Then set the lens on the camera to
    infinity focus. That will produce a good quality collimated source.

    However, looking at the distances you used, the light is pretty
    collimated over the extent of the frame - 7m range on a 28.7mm wide
    frame is only a divergence of 4mrad, even if you consider the entire
    aperture of the lens mount, it is a divergence of less than 8mrad.
    Provided you can avoid scattering centres in front of the camera you
    will be hard pressed to find an incoherent collimated source that is
    much better, without specialist kit.
    I am getting consistent results which are much less than cos(a) - see
    below. Unless I just have a weird sensor, it seems clear that Canon are
    doing something with their full frame sensors to counteract this problem
    before it arises.
    Yes - it is amazing at spotting dust! That huge effective f/# helps,
    although you have to be careful not to get drawn in by how bad it looks.
    I noticed a couple of huge hairs that always seemed to be on the edge of
    the shadow at negative azimuth angles, no matter what angle the camera
    was at. It took me a little while to realise these were particles of
    dust actually on the edge of the lens flange itself! They look pretty
    sharp, even that far from the focal plane. ;-)

    Anyway, as promised, tonight I tried to address a wider range of angles
    and eliminate some of the possible issues that people had raised in the
    first attempt. I still used the white LED source, but sited the camera
    4.5m away. This gives an effective f/900 beam, so virtually no
    variation in light intensity across the frame. This larger distance
    meant a longer shutter time, of about 1.5sec, for the same exposure, but
    I increased this to 2.5sec so that the numbers would be larger and thus
    emphasise any variation that was present. I also shot all of the images
    in RAW format with 10 frames at each angle of incidence to check for
    consistency. The response was averaged over a block of 300x400 pixels
    in the image, approximately 200pixels out of the shadow for consistency,
    although the +30 elevation measurement was right against the shadow due
    to the light just barely reaching the edge of the frame. The average
    standard deviation across any 10 frames at the same angle of incidence
    was 0.04%, the worst was 0.08% at one of the extreme azimuth angles,
    whilst the best was on axis with a standard deviation of 0.002%. These
    figure show the results are very consistent and reproducible.

    I measured the response relative to on axis for azimuth angles in 10deg
    steps from -40deg to +40deg. To achieve +/-40 angle of incidence
    requires the light to enter the camera at one edge of the lens mount and
    merely fit into the far edge of the frame - way beyond what any real
    lens could ever achieve in practice. I also measured some spot
    responses at +/-30deg angles of elevation, to see how that faired. As
    you can probably tell from looking at the Canon lens mount, the
    electrical contacts for the lens obscure the +ve angle of elevation more
    than the -ve angles, which meant that only the 0deg azimuth measurement
    could be obtained at +30deg elevation. Unfortunately, my tripod is only
    marked off accurately in +30deg steps of elevation, so although I could
    have got finer angles they would have been more guesswork than

    So, to the results:

    Response as a percentage of 0az, 0el.
    Azimuth Elevation (deg)
    (deg) -30 0 +30
    -40 x 86.76 x
    -30 94.48 97.01 x
    -20 x 97.55 x
    -10 x 98.69 x
    0 98.82 100.00 95.19
    +10 x 99.66 x
    +20 x 98.94 x
    +30 95.29 96.5 x
    +40 x 86.86 x

    As you can see, this is very flat response, with no appreciable fall off
    across the entire range. Even out to +/-30deg azimuth, there is only
    3.5% loss - similar to my previous measurement - although there is then
    a more substantial drop at +/-40deg, with approximately 13% loss on
    either side. As previously mentioned these extreme angles are really
    grazing inside the lens mount and just reaching the far side of the
    frame - so totally impossible for any real lens to achieve.

    The response is equally flat with elevation angle, possibly showing a
    slight bias to the -ve elevations with less than 1.2% loss at (0,-30)deg
    however, as already mentioned, the +30el measurement was tight between
    the frame edge and the edge of the shadow area, and this may have caused
    some underestimate of the response.

    The gradually increasing roll off with angle of incidence is also
    confirmed by the two measurements at (+/-30, -30) az,el. This
    combination produces a total angle of incidence of approximately 42deg
    in the diagonal direction relative to the pixel matrix, and is
    significantly better than the response at +/-40az - so the angular
    response is *not* circularly symmetric but has a clear x-y structure to

    Nevertheless, these figures (even including the +/-40 deg values) are
    significantly *less* than would be predicted by a normal cosine law -
    30deg would be expected to be 13.4% down, with 40deg about 23.4% down in

    So, contrary to all of the claims about the Canon 5D sensor being worse
    than film in its response to angle of incidence, it actually appears to
    be significantly better and will, as a consequence, produce less light
    fall-off with super-wide-angle lenses than film. Nevertheless, the
    difference between the angular response of the 5D sensor and the
    prediction for a flat sensor surface such as film emulsion is
    insignificant compared to the typical vignette produced by such lenses
    themselves on *either* medium. In short, there is no light fall of at
    the edges and corners of the 5D sensor - to all intents and purposes the
    light fall off is *ALL* caused by the lens, *not* the sensor.

    To those who still protest that this doesn't test what a lens will
    produce - don't bother replying here. Your time would be much more
    fruitfully served learning about basic optics than either protesting
    about the measurements or your imagined worse performance of the 5D than

    I would, however, like to see some other FF results - because there is a
    clear difference between these measurements and the anonymous 10DMkII
    owner. It would be interesting to know if this is consistent across the
    FF range, a new feature of the 5D sensor or a fluke sensor in my camera.
    If its the latter, and anyone from Canon is reading, I'll gladly swap it
    for a 1DsMkII or better, if you want it back for analysis. ;-)

    As mention in the opening post of this thread, I would also like to hear
    from Olympus owners if the ultra high effective f/# test shows any dust
    present on the sensor that is not otherwise present in images.
    Kennedy McEwen, Mar 18, 2006
    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.