What compact digicam has the biggest CCD pixels?

Discussion in 'Digital Photography' started by Paul Rubin, Apr 22, 2005.

  1. Well, 4 stops is extreme. But there is certainly not a 16x difference
    in signal to noise ratio. Perhaps 2 stops of real loss (see below).
    18 or 20 bit converters are totally unnecessary with current
    high end sensors. The 1D Mark II has a full well of about
    52,000 electrons. That requires 15.7 bits (16 bits) to digitize
    every electron (if that could be done). With read noise
    of about 7 electrons, one needs to digitize to about half
    the read noise, or 3.5 electrons. 52000/3.5 ~14900 which
    needs 13.9 bits. A 14 bit converter pretty much does it.

    This also gets back to your above example of 4 stops.
    A 12-bit converter certainly loses about 2 stops of information
    that is really there in the signal.

    Roger N. Clark (change username to rnclark), May 6, 2005
    1. Advertisements

  2. Paul Rubin

    Paul Rubin Guest

    Apparently Sony uses 14 bit converters. I wonder why Canon and Nikon
    don't. And Sony uses ccd's with much smaller pixels, where 14 bits is
    really too much.
    Paul Rubin, May 6, 2005
    1. Advertisements

  3. Paul Rubin

    JPS Guest

    In message <>,
    I don't have any special problem with my 20D. It is much better than
    most of the "problem" 20Ds that people have posted samples from. I
    normally don't get any kind of visible banded noise unless the light
    source is extremely unbalanced (just after sunset, for example, when the
    light from the sun is coming down diffused and lacking in red, or under
    tungsten lighting, which lacks blue), with a bit of general
    under-exposure. The problem is, you can clearly see that the sensor is
    doing a much better job than the A2D is, and it is annoying to know that
    the sensor is very sensitive, but you won't see what it captures,
    because of the digitization hardware.

    I know you're probably thinking about the RAW value of a single
    electron, but I think you are missing the forest for the trees. People
    don't take photographs to obtain 99%-accurate values for each pixel, as
    a measuring intrument! They take photographs to capture *images*, and
    even when single pixels start to fail you, average pixel values for a
    small area contain more accurate recordings at lower spatial

    Many people have sent their cameras back to Canon because of banding,
    and Canon has returned them, saying that they are OK, without fixing
    anything. They deny a problem, even when it is pronounced.
    The best way is to remove the banding from the RAW data itself. The
    red, green, and blue channels all have different levels, locally, but
    the green-filtered sensels represent 1/2 of each row and column of
    sensels, so blackpoint adjustments of all sensel data on each row or
    column based only on the green is probably the best route to take.
    Generic noise reductions are relatively ineffective, as the problem
    isn't really noise itself, but offsets by row and column, combined with
    posterization of low RAW values. There are still blackpoint offsets
    after the noise is removed.
    JPS, May 6, 2005
  4. Paul Rubin

    JPS Guest

    In message <>,
    Maybe they don't use variable gain, but are simply bit-shifting for ISOs
    200 and 400? That might mean that a good deal of the noise at ISO 400
    comes from posterization.
    JPS, May 6, 2005
  5. [A complimentary Cc of this posting was sent to
    Roger N. Clark (change username to rnclark)
    I see: as Alice expressed it, "words should mean the things I want them to

    As any Poisson noise, photon Poisson noise is easily expressed via the
    count of photons. And given that the count of photons is different
    than the count of "converted photons", "converted photons" Poisson
    noise is very different from *photon* Poisson noise.

    But it is very easy to avoid this confusion: just replace all the
    occurences of "photon Poisson noise" on your website with "electron
    Poisson noise". (I do not think I saw any occurence which did not
    mean "electron Poisson noise".)
    So you continue putting words in my mouth... Continuing with this
    logic, next time you will insist that I say that

    "These devices do not count anything, they just produce some
    electric signals through the RS232 (or whatever) interface".

    If you calibrate your termometer so that its scale is the number of
    incoming photons, then (in certain conditions) it may be used as a
    photon counting device. 99% of measurement is calibration; the
    internal working of the measuring device is an ancillary topic.

    AND: this has no relationship to the fact that

    photon Poisson noise != converted photon Poisson noise.
    The principal question, from my POV, is not what is "standard", but
    what is *confusing*. Your usage of "photon Poisson noise" *is*
    confusing. (I doubt that it is standard, but this is a *different*
    Do not think so. If not in "Alice mode", then you need *calibration*;
    i.e., something like

    "incoming photons" = "counted photons"/QE.
    As I said it many many times already: I can agree *in one case only*:
    when you consider the noise of the whole system: "incoming photon
    noise" + "noise GENERATED by the sensor due to its low QE". But
    oftentimes you do the same in different situations.

    *This* is the distinction I talk about "the sensor" vs "a sensor". A
    *different* sensor will have different generated noise, thus different
    total noise (in the same situation). When you discuss one particular
    sensor, then combining two different sources of noise together makes
    sense (in some situations); if you are not confined to using one
    particular sensor, one should consider two types of noises (either
    incoming and generated, or incoming and total) separately.


    Ilya Zakharevich, May 8, 2005
  6. Try searching for photon counting systems or, for example,
    "1p21 photon counting" The 1P21 photomultiplier tube
    was used for decades (and still is) for photon counting even though
    its quantum efficiency is very low (on the order of a couple of percent).
    Astronomical Photoelectric Photometry Manual
    (read the introduction).

    Photon counting devices from Hamamatsu Corporation:

    Search: photon counting detectors

    Determining the absolute responsivity of photon counting detectors

    Edinburgh instruments offer a range of detector modules that are suitable for
    either single photon counting applications or analogue detection:

    Now try searching on: photon counting Poisson statistics
    google gives about 46,700 hits.

    Now try searching on photon counting Poisson statistics CCD
    Google gives 7,050 hits

    read www.phys.vt.edu/~jhs/phys3154/snr20040108.pdf
    .... measurements are made by counting the number of electrons released by the
    incident photons. ... in the CCD chip also obey Poisson statistics. ...

    Variance vs Mean Photon Count from a CCD Detector:

    Thousands of other pages to read.
    Roger N. Clark (change username to rnclark), May 15, 2005
  7. [A complimentary Cc of this posting was sent to
    Roger N. Clark (change username to rnclark)

    BTW, I see now that "photon Poisson noise" is indeed a "standard" term
    nowadays for what should be named "electron noise". Which is very sad
    - especially if you take into account that the stats of this noise are
    not even Poisson! (Though the stats become very close to Poisson when
    the number of incoming photons is "large enough".)

    Ilya Zakharevich, May 16, 2005
  8. [A complimentary Cc of this posting was NOT [per weedlist] sent to
    Ilya Zakharevich
    Drat, the statement that this noise is not exactly Poisson is most
    probably wrong - if no incoming photon produces more than 1 electron,
    the noise is exactly Poisson (though, obviously, with smaller S/N
    ratio than the incoming flow of photons). Actually, some time ago I
    actually wrote a correct treatement of this subject, but now I took a

    Here is the whole picture: consider an incoming flow of objects (in
    our settings, photons) entering a black box; suppose that the black
    box ("sensor", or "amplifier") produces several "outgoing objects" (in
    our settings, electrons) for each incoming object; there are some
    probabilities p0, p1, p2 etc of producing 0, or 1, or 2, etc. outgoing
    objects per an incoming one.

    Assume that the statistics of the incoming flow is Poisson; when is
    the output statistic Poisson too? Answer: exactly when only 0 or 1
    objects may be produced; in other words, when p2 = p3 = p4 = ... = 0.

    So if it is "an amplifier", the output is not *exactly* Poisson (but
    may be quite close to a Poisson - similar to Poisson being close to
    normal); but certain types of "attenuators" will produce Poisson output.

    Sorry for the confusion,
    Ilya Zakharevich, May 20, 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.