supplierdeeply

Canon noise reduction

  • 25 Replies
  • 6760 Views
  • Publish
    Canon noise reduction
    « on: 07 / May / 2008, 14:22:36 »
    Advertisements
    Hi,
    The noise reduction options on the raw parameters menu page are mis-named.  They should be called dark frame subtraction.  That is because Canon also does other noise reduction that this option does not affect.  For increasing ISO Canon applies more noise reduction until it gets very strong at ISO 1600 on my A720IS.  As far as I know it may do some noise reduction even at ISO 80, but I can't tell that.  I can tell that it is doing noise reduction (mainly to chroma (color) noise) at ISO 400-1600.  I think it is doing something at 200, and maybe even something at 100.  At 800 and 1600 the noise reduction becomes very pronounced.

    This noise reduction gets applied during raw development.  You can experiment with it by saving a raw and developing it in-camera at different ISO settings.  The effect is subtle since it is mainly being applied to colors and not to brightness.  We humans don't see color very well and so smoothing colors is much harder to detect than smoothing brightness.  That is why Canon includes one set of color info yet four sets of brightness info for every two pixels on the viewport.

    Jon
    « Last Edit: 07 / May / 2008, 17:16:39 by hiker_jon »

  • Publish
    Re: Canon noise reduction
    « Reply #1 on: 07 / May / 2008, 16:25:53 »
    that is why Canon includes one set of color info yet four sets of brightness info for every two pixels on the viewport.


    So, what does two sets of brightness info per pixel mean ?   :)


    David

  • Publish
    Re: Canon noise reduction
    « Reply #2 on: 07 / May / 2008, 17:14:47 »
    Quote from: Microfunguy link=topic=1286.msg11735#msg11735
    So, what does two sets of brightness info per pixel mean ? 

    Investigation with a magnifying glass shows that the screen on my A720 consists of 240 lines each with 180 sets of R,B,G dots.  The lines are offset by half a dot width and even lines start with blue and odd lines start with green.  Thus the screen is a complex pattern of dots.  I think the extra brightness info is used to anti-alias the screen so that aliasing patterns (say in a picket fence) are not so obvious.  The "advertised" size of the screen is 320x240.  The size of the buffer is 360x240x3bytes.  Maybe that should be thought of as 180x240x6bytes.  The 6 bytes hold info for 4 pixels.

    Jon
    « Last Edit: 09 / May / 2008, 16:43:37 by hiker_jon »

  • Publish
    Re: Canon noise reduction
    « Reply #3 on: 09 / May / 2008, 11:54:02 »
    Noise Calculations:  I took some raw data ,developed it at different ISO's, and then calculated the standard deviation of a small rectangular area. I did this in YCbCr color space. I did it separately for Y (set Cb,Cr to 128) and CbCr (set Y=128).  This is not Canon's YUV colorspace, but it is probably close enough to see what is going on. 

    My photo was not of uniform areas of color as it should be for the best analysis of the noise reduction.  It was bluish gray with some reds.   There was structure in both luminance and colors in the image.  This means that the SD in an area reflects both the structure in the image and the noise.  If I ever have a lot of time I will repeat the experiment with uniform areas of color and gray.

    Here are the results:

    ISO     YSD     CbCrSD
    80       22.3     1.23
    100     22.3     1.22
    200     21.2     1.22
    400     20.5      .82
    800     20.1      .76
    1600   22.8      .62  ***

    Conclusion: No noise reduction at 100, a tiny bit at 200, heavy color noise reduction at 400, increasing at 800, increasing at 1600.

    ***The +1 EV in 1600 development multiplied the image data by 2.  So the SD should have been doubled if there were no noise reduction.  Thus there is very significant noise reduction for 1600.


    *

    Offline fbonomi

    • ****
    • 469
    • A570IS SD1100/Ixus80
      • Francesco Bonomi
  • Publish
    Re: Canon noise reduction
    « Reply #4 on: 09 / May / 2008, 13:44:56 »
    maybe a dumb quesyion but ... are you saying that the ISO is applied olny AFTER the RAW is captured?

    I mean, that when saving RAW ISO settings are completely irrelevant (a RAW shot as ISO 80 is identical to a RAW shot at ISO 1600)?

    I thought there would be some "amplifiers" of sort increasing the electric signal from sensors...


    *

    Offline BB

    • ***
    • 164
  • Publish
    Re: Canon noise reduction
    « Reply #5 on: 09 / May / 2008, 14:00:04 »
    Somebody else here (hiker_jon) found that for the A720, the RAW at ISO 1,600 and ISO 800 where exactly the same levels... So, in the case of the A720, it appears that the gain from 800 to 1,600 is just a multiplier applied to the 800 RAW data (plus more noise smoothing too?).

    in-camera EV+1


    -Bill

  • Publish
    Re: Canon noise reduction
    « Reply #6 on: 09 / May / 2008, 15:17:58 »
    I mean, that when saving RAW ISO settings are completely irrelevant (a RAW shot as ISO 80 is identical to a RAW shot at ISO 1600)?
    For 80 to 800 the raw is different at each ISO.  The gain of the chip is indeed turned up.  For 1600 I guess the increase in gain is too much for the chip.  So rather than increasing the gain Canon uses the same gain as 800 and develops the data differently to achieve EV+1.

    The development of the data is different for different ISO's.  In my experiment I used a raw taken at ISO 80 and developed it with the camera set to different ISO's.  I have shown that for ISO 200, 400, 800 and 1600 there is different noise reduction depending on the camera ISO setting.

    So if you save raws and your picture taken at ISO 200 is too noisy you can use CHDK to develop the raw at ISO 800.  Conversely if you have an image at ISO 800 that has colors too smoothed by the noise reduction you can develop at ISO 80 to eliminate the noise reduction. 

    Thus CHDK gives adjustable noise reduction to Canon cameras.
    « Last Edit: 09 / May / 2008, 15:28:05 by hiker_jon »

    *

    Offline fbonomi

    • ****
    • 469
    • A570IS SD1100/Ixus80
      • Francesco Bonomi
  • Publish
    Re: Canon noise reduction
    « Reply #7 on: 09 / May / 2008, 15:36:11 »
    ok, so there is a hardware and a software component.
    thanks, good to know.
    I am studying the raw data sensors to determine luminance of the scene, and I was ... puzzled at the thought that I could hae everything wrong...

    I just have something wrong :-)


    *

    Offline fudgey

    • *****
    • 1705
    • a570is
  • Publish
    Re: Canon noise reduction
    « Reply #8 on: 09 / May / 2008, 18:41:13 »
    A nice find... but are you sure your data is accurate enough to confirm there is NR happening at ISO 200? What I'm thinking is burst mode speed. It gets slowed down significantly if ISO is increased. I figured this software NR would be the reason, but for what I can remembere, at least on my camera ISO200 burst is still fast and ISO400 slow.


  • Publish
    Re: Canon noise reduction
    « Reply #9 on: 09 / May / 2008, 22:40:34 »
    A nice find... but are you sure your data is accurate enough to confirm there is NR happening at ISO 200? What I'm thinking is burst mode speed. It gets slowed down significantly if ISO is increased. I figured this software NR would be the reason, but for what I can remembere, at least on my camera ISO200 burst is still fast and ISO400 slow.
    The SD is only very slightly reduced, this may be experimental error.

     

    Related Topics