Light fall off on dSLRs - an experiment

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

  1. Following from the discussion here over the past few days about light
    fall off on the Canon 5D and whether this is worse that it was with film
    I decided to try a little experiment.

    The first step was to eliminate any effect of light fall-off caused by
    the lens, which was quite simple - eliminate the lens! ;-)

    So I set up 20mA current flowing through a 5mm AGI-5N3CMPW white LED
    (data sheet at to produce
    a "point source of light".

    The Canon 5D camera without lens was mounted on a tripod about 80cm from
    this LED source in a darkroom, so the only light reaching the camera was
    from the LED. With the camera facing the LED, the TTL meter indicated a
    manual exposure of 1/80th second.

    Two exposures were made in jpeg format at this exposure. The first with
    the camera facing the light source, giving light perfectly perpendicular
    to the sensor. The second with the camera rotated on the tripod head so
    that the light source cast a shadow across the centre of the frame - the
    most extreme angle of incidence that it is possible to create from any
    lens through the Canon mount. In fact, since the lens requires a
    physical mount, which takes up some space, this is actually a steeper
    angle of incidence than would be possible with a real lens. This was
    less simple to achieve, since the focus screen diffuses the lens mount
    shadow quite a lot, so a couple of test shots were made and reviewed on
    the LCD screen to get the shadow in the centre of the frame.

    The two jpeg images were then imported into Photoshop and cropped to the
    exposed areas, obviously only half of the frame was exposed on the
    extreme angle, with the lens mount casting a shadow over the other half.
    Histograms for the exposed areas were:

    Central source: 73.80 Photoshop levels.
    Edge source: 71.82 Photoshop levels.

    This indicates a light fall-off due to extreme angle of incidence of
    only 2.68%, or approximately 4 HUNDREDTHS of a stop!

    This simple test, which *anyone* can independently repeat with their own
    camera and confirm for themselves, categorically *PROVES* that there is
    essentially *NO* sensitivity to angle of incidence on the Canon 5D
    sensor (and I suspect *any* dSLR sensor!).

    Any light fall off that is present on this camera is, to all intents and
    purposes, *EXACTLY* the same as it was for full frame film cameras - the
    effect is *ALL* in the lens.

    This simple test, which *anyone* can independently repeat with their own
    camera and confirm for themselves, also completely *debunks* one of the
    primary Olympus argument in favour of the 4/3 format!

    By the way, this is *NOT* a test for those of a nervous disposition
    likely to be scared by seeing dust on their focal plane. A 5mm source
    at 80cm range is equivalent to shooting at f/160 and every single spec
    shows up. The test images from my apparently clean 5D sensor looked
    positively filthy! ;-) I would, however, be interested to hear from
    any Olympus owners who try it, since it would show whether the
    ultrasonic cleaner really does work.
    Kennedy McEwen, Mar 16, 2006
    1. Advertisements

  2. Kennedy McEwen

    Jeremy Nixon Guest

    Not a valid test, since it's the lens that causes extreme angle of
    incidence of light on the sensor.
    Jeremy Nixon, Mar 17, 2006
    1. Advertisements

  3. Kennedy McEwen

    eawckyegcy Guest

    Don't be an idiot: the sensor doesn't know (let alone care) where the
    photons are coming from.
    eawckyegcy, Mar 17, 2006
  4. Kennedy McEwen

    Jeremy Nixon Guest

    So you're saying that the light falling on the sensor is the same with or
    without a lens? That's just not the case.
    Jeremy Nixon, Mar 17, 2006
  5. Kennedy McEwen

    eawckyegcy Guest

    One of the idiot theories being bandied about here is that there is
    something special about a digital sensor in that it can't handle light
    "landing" on the pixel at a high angle of incidence.

    Look at the diagram. The pixel has no idea where that photon is coming
    from, beyond its angle of arrival. None. It matters zero whether the
    last thing the photon touched was a piece of glass or the PN junction
    in an LED.

    Basically, McEwen has shown the reigning (idiot) theory is, as
    expected, total bullshit. His simple demonstration essentially makes
    most of the Nikon Nutcases here look like ignoramuses, and if they
    persist in the face of physical reality, ineducable idiots.

    I know you are smart Jeremy. Please make the right choice.
    eawckyegcy, Mar 17, 2006
  6. The light source is at an even more extreme angle of incidence than it
    is possible for any practical lens to produce! If it doesn't show up on
    this test then it isn't going to be produced by a lens.
    Kennedy McEwen, Mar 17, 2006
  7. What makes you think the sensor treats light from a lens any differently
    from any other source of light? If the detector has less sensitivity to
    light from the lens at extreme angles then it has less sensitivity to
    light from any source at this same angle. What this experiment does is
    essentially two fold:
    1. It eliminates vignetting effects present in the lens itself, which
    would cause corner fading on all sensors whether film or digital
    2. It provides a direct comparison between light at the most extreme
    angle of incidence compared to light that is incident perpendicular to
    the focal plane.

    Light fall-off is not caused by the sensor.
    Kennedy McEwen, Mar 17, 2006
  8. Kennedy McEwen

    Tony Polson Guest

    It's a pity you went to all that trouble to carry out a completely
    irrelevant and grossly misleading test.

    To be of any worth, your point source should have been somewhere close
    to the distance from the sensor represented by the rear element of a
    lens such as an EF 24mm or 28mm. Only then would you have seen the
    significant difference in illumination caused by the changed angle of
    incidence - you would also have to make an allowance for the decreased
    illumination of the corners due to the inverse square law, but the
    light fall off due to incident angle to the sensor would still have
    been clearly apparent.

    So what did you do instead? You put the light source 80cm away so the
    angle of incidence at the corner was only 89.93 degrees rather than
    90.00 degrees at the centre. 0.07 degrees different? It's so small,
    it is almost impossible to measure. No wonder there was no difference
    in illumination!

    What a complete waste of time. You have proved nothing - except your
    own complete inability to understand the problem.
    Tony Polson, Mar 17, 2006
  9. Kennedy McEwen

    Jeremy Nixon Guest

    I don't know sensor design. But people who do say differently, especially
    with the Bayer filter and microlenses and whatnot involved.
    "Nikon Nutcases" is a pretty silly indictment. I have nothing against Canon.
    I see no real advantage in 35mm sensors, from Canon or Nikon or anyone else.
    Jeremy Nixon, Mar 17, 2006
  10. Kennedy McEwen

    Battleax Guest

    Don't be an idiot
    idiot theories
    reigning (idiot)
    total bullshit
    Nikon Nutcases
    ineducable idiots

    Wow, a most intelligent presentation.
    Battleax, Mar 17, 2006
  11. Kennedy McEwen

    bjw Guest

    First, let me say that I take no brand side in these DSLR battles
    and I think this is a big non-issue and people should go make
    pictures rather than flaming each other over spec-sheet
    hypotheticals. Also I am in favor of measuring things rather
    than theorizing in a vacuum.

    That said, I don't understand how you could get the same light
    level in this experiment. The LED is at the same distance
    from the sensor in both cases. But in the second case, the
    camera is rotated, so each pixel surface is at an angle to the
    LED. Put another way, from the LED's position, each pixel
    subtends a smaller angle when rotated, so it should receive a
    smaller amount of light. The amount of light falling on a
    given pixel should be smaller by cos(theta), where theta is
    the angle by which you rotated the camera. This is
    independent of any arguments about whether or not the
    pixel _sensitivity_ is dependent on angle of incidence - it
    would be equally true of film.

    Assuming that the Canon sensor to mount distance is 44mm
    and the diameter of the lens mount is 54mm, you would have
    rotated the camera by about theta = arctan(27/44) = 31.5 deg
    to put the lensmount shadow on the center of the sensor.
    Then the cos(theta) factor should be 0.85, so the signal
    should be 15% lower even if the pixels are completely
    insensitive to angle of incidence.

    I don't know enough about how the in camera processing
    determines the output jpeg pixel values. Is it possible that you
    need to shoot raw to do this test?
    bjw, Mar 17, 2006
  12. Kennedy McEwen

    eawckyegcy Guest

    He put the source 0.8m away and rotated the sensor. I guess you missed
    that part, eh?
    You are beyond all possible hope, Polson.
    eawckyegcy, Mar 17, 2006
  13. Kennedy McEwen

    Jeremy Nixon Guest

    It's not a matter of the sensor treating it differently. The light is
    different coming from a lens than what you've tested.
    Your experiment doesn't test this. Your light source was so far away from
    the sensor that has no real relevance to what happens when a lens is on the
    Jeremy Nixon, Mar 17, 2006
  14. Kennedy McEwen

    w.beckley Guest

    Isn't it? The lens is going to reduce transmission, and it is going to
    focus the photons, directing them from the scene onto very specific
    parts of the sensor at angles ranging from 0 to (perhaps) 90 degrees.
    But none of tht changes the nature of the light for the purposes of
    this test. This test is merely determining what the exposure loss is as
    a result of the angle of the light hitting the microlenses. And this
    suggests that it is very little at all.

    Does the lens matter? Yes, obviously. A lens will vignette, and this
    test doesn't refute that. It merely suggests that the vignette will be
    similar to what was seen on film.

    I'd like to see the test shots for myself, but the test itself seems
    pretty strongly conceived.

    w.beckley, Mar 17, 2006
  15. People with a vested interest say things which are not entirely
    accurate, that is why independently verifiable tests and measurements
    Kennedy McEwen, Mar 17, 2006
  16. Kennedy McEwen

    eawckyegcy Guest

    Don't like what I write, killfile me. If you identify the newsreader
    you are using, I'll happily google up the instructions for you.
    eawckyegcy, Mar 17, 2006
  17. In what way is this light different - do all the photons carry
    individual "I came from the lens" ID tags?
    The reason the light source was placed so far away was so that the angle
    of incidence could be changed by one, and only one, adjustment -
    changing the angle of focal plane relative to the light source. What
    other variation would you suggest is necessary?
    Kennedy McEwen, Mar 17, 2006
  18. In the Canon lens mount, this angle cannot exceed about 26deg, due to
    the need for the mirror to clear the rear lens element and the limited
    diameter of the lens mount.
    Kennedy McEwen, Mar 17, 2006
  19. Kennedy McEwen

    Jeremy Nixon Guest

    A pixel on the sensor is not illuminated by a point light source. It's
    a circular light source. From the perspective of the edges of the sensor
    it is an elliptical source. You can't model this by envisioning a single
    ray coming from a point in the scene and landing on the corresponding
    point on the sensor, because that's just not how it works, a fact you can
    easily demonstrate by, for example, noting the effect that blocking a
    small portion of the lens has on the final image.
    Jeremy Nixon, Mar 17, 2006
  20. Kennedy McEwen

    eawckyegcy Guest

    They can say what they like, but the data remains.
    eawckyegcy, Mar 17, 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.