A question about ISO on a hacked Canon dRebel

Discussion in 'Canon' started by Al Dykes, Dec 20, 2004.

  1. Al Dykes

    Al Dykes Guest

    I've loaded the hacked code into my dRebel and it works fine, so far.

    When I set ISO to 3200, is this done by bit shifting or analog
    amplification ?

    In some usenet group someone said that 800 _might_ be the fastest
    native speed of the sensor, and that 1600 was done by shifting bits,
    not by amplification.

    I'm sure I don't understand the pros and cons of either aproach.
    Amplification may just increase noise, shifting is at th eexpense of
    dynamic range, and still picks up noise. Right ?

    Comments ?
    Al Dykes, Dec 20, 2004
    1. Advertisements

  2. I think it's a shift left which doubles the binary value and that would
    equal one f-stop. I have a 300D (not hacked) and have not found ISO 1600 to
    be all that useful. It's noisy and ISO 3200 would be worse. I also know
    that some DRebel users like ISO 1600 and 3200 and get decent results with
    noise reduction software ... just not to my liking.
    Charles Schuler, Dec 20, 2004
    1. Advertisements

  3. Al Dykes

    Alan Browne Guest

    Pretty much. I don't know how many stops of pre-A/D gain there are. Bit
    shifting (up) will 'noisify' shaddow detail and put high dynamic detail at risk
    of obliteration. You will get more quantization noise (the pattern created by
    unfiltered digital amplification) which may make smooth/dark areas look lumpy.

    Beyond the mechanics above, the s/w in the camera is also optimizing its
    conversions from the sensor reading when converting to JPG (and in the PC when
    converting from RAW to TIFF/JPG, etc.) These are very privy to the OEM.
    Minolta for example have something in the camera called: "CxProcess™ III Image
    Optimization" and there's no telling specifically what goes on in there...
    analogous signal processing takes place in pretty much all digital cameras.

    From the Minolta site: "Konica Minolta’s exclusive CxProcess III technology
    brings out the best from the Maxxum 7D’s 6.1-million effective pixel APS-C size
    CCD. CxProcess III optimizes color saturation, edge sharpness, and
    highlight/shadow contrast to ensure that colors are rendered true-to-life.
    CxProcess III also suppresses noise during slow-shutter imaging, and assures
    rich textural detail and spatial perspective for vivid yet natural reproduction
    of skin tones, clouds, and other subjects with delicate surface textures."

    Alan Browne, Dec 20, 2004
  4. Al Dykes

    Ryadia Guest

    If you ever find yourself in one of those rare instances when you get the
    Minolta and a pair of Canon (20D and 1D,II) DSLRs together at the same time
    to compare the quality of output, you might discover a major contridiction
    in the quoted text.
    Ryadia, Dec 20, 2004
  5. Al Dykes

    Don Farias Guest

    You might want to look at the D70 review in Dpreview where there are side
    by side comparisons of the resulting noise output between the D70 and the
    300D (with normal firmware) are displayed. The "character" of the noise at
    1600 is described as "blotchy" vs. more fine grained for the D70.
    This *might* confirm the comment made about the 800 native speed
    Don F
    Don Farias, Dec 22, 2004
  6. Al Dykes

    JPS Guest

    In message <>,
    One reason why quality nose-dives at the higher ISOs is not just the
    noise itself; higher ISOs are often used in environments with poor
    *quality* lighting, not just low quantity lighting. There tends to be
    more shadows and highlights in indoor lighting, especially when the
    walls, floors and ceilings are dark, or spotlights are used.

    If you shoot ISO 1600 outdoors with good lighting, you might not even
    notice the noise until close inspection.
    JPS, Dec 29, 2004
  7. Al Dykes

    JPS Guest

    In message <cq6udm$sbb$>,
    3200 uses shifting.
    That may have been me. I remember getting the same values from my 10D
    in the RAW data that I got for ISO 3200. I have better means of
    verifying this now, and will give more exacting details. I was going by
    DCRAW's output, which is kind of funky. Now, you can take a test shot
    in RAW mode, convert it to an uncompressed Adobe DNG, and look at the
    RAW data in all its rawness.
    I suppose that at the highest amplifications, amplification artifacts
    become so strong that one stop of amplification might be worse than one
    stop of quantization, so above ISO 800 or 1600 they switch to the
    shifting, to make the "experience" of using different ISOs a smooth
    transition. I don't know why they didn't use intermediate values other
    than powers of two; say an amplification of 24x and multiply the numbers
    by 1.333. Maybe that causes banding in the shadows, or the computation
    takes too much time, but if the numbers were left alone, it could be
    compensated better in a RAW converter where there would be no
    intermediate posterized data.

    Personally, I would like a custom function that turned on pure
    amplification mode, and let *me* do the under-exposing, including
    intermediate ISOs above 800, where things tend to change fast.


    I just took blackframe and super-overexposed white wall images at ISOs
    800, 1600, and 3200 on my 10D. I converted them all to uncompressed DNG
    files, and looked at the RAW data in a hex editor (set to look at the
    data as decimal numbers, assuming 16-bit unsigned data). The data
    patterns are the same for the 1600 and 3200, and both are a little
    strange. You get a long string of even numbers, then a long string of
    perfectly alternating odd and even numbers, then a long string of odd
    numbers, then a long string of alternating numbers again. I don't know
    if it's the camera or the DNG converter that is doing this to the data
    (adding or subtracting one to blocks and striped blocks), but it's quite
    clear that there are only 11 bits used for both ISO 3200 *AND* ISO 1600.
    Here is some sample data from the ISO 1600 blackframe:


    Note how only the vertical stripes marked have odd numbers.

    JPS, Dec 29, 2004
    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.