Noise Test

Discussion in 'Digital Cameras' started by Dudley Hanks, Feb 6, 2014.

  1. Dudley Hanks

    Dudley Hanks Guest

    Out of curiosity, I took my XSi and shot a pic at 1/4000 sec,
    f/22 and ISO 100, with the lens cap on, in a dark

    My assumption is that any pixels not registering pure black can
    be attributed to noise.

    I observed two interesting points:

    1. The noise level was a lot higher than I anticipated,

    2. The noise level was worse when a JPEG was analyzed than when
    a PNG was used. (Both test images were obtained from the same
    CR2 file.)

    of the 12 meg pixels, almost a quarter registered higher than 0
    for the JPEG conversion, and roughly 1/8 when a PNG was

    Using a 256 gray scale, the highest value for the JPEG was 12, 13
    for the PNG image.

    I'm wondering if anybody out there has done any similar tests.
    And, if so, what results were obtained for their

    Take Care,
    Dudley Hanks, Feb 6, 2014
    1. Advertisements

  2. Dudley Hanks

    Me Guest

    I haven't done any tests, but:
    png uses lossless compression (decompress a PNG and you get exactly the
    same as original uncompressed file) - unlike jpeg, which would
    (depending on compression setting etc) average/smooth out individual
    bright pixels into larger blocks, hence the higher number of >0 pixels.
    You really see this "blockiness" in highly (over)compressed jpegs.

    The noise is at dark levels which are generally of no consequence to
    photography - as there's (usually) no useful image data way down there.
    An exception to that is for multiple exposure / image stacking, where
    the random noise can be subtracted out by averaging, and faint but
    "real" data recovered.
    Me, Feb 6, 2014
    1. Advertisements

  3. Dudley Hanks

    Dudley Hanks Guest

    When I took the pics, I was in a room with a lot of
    electrically powered devices, and I'm wondering if the noise
    can be attributed to interference from these devices.

    As soon as I get some time, I'm hoping to run some tests by
    placing the camera next to various devices while shooting to
    see if different results are obtained.

    Any idea whether this sort of noise originates within the cam
    or from exterior influences?

    Take Care,
    Dudley Hanks, Feb 6, 2014
  4. It is called "read noise", and is inherent in the
    electronics of the camera.

    At ISO 100 there is little or no analog amplification of
    the sensor's signal, but for ISO's higher 400 there is.
    I don't know for sure how your particular camera does
    it, but most likely the differences in ISO's from 200 to
    at least 1600 and maybe much higher are the result of
    analog amplification of the sensor signal.

    The significance will probably be that increasing ISO
    will cause an increase in noise levels. However,
    different RAW processing can easily hide that, so you
    won't necessarily see it when looking at anything other
    than RAW sensor data.

    In fact, there isn't much point in experimenting with
    what you are seeing in a JPEG image, simply because what
    you can see depends so much on how it is processed and
    so little on what the camera produced.
    Floyd L. Davidson, Feb 6, 2014
  5. Dudley Hanks

    Dudley Hanks Guest

    I limited the test to ISO 100 because I thought that would
    give me the truest representation of how much noise is
    contained in an image.

    In future tests, though, I may try to get an idea of how much
    impact increasing the ISO produces.

    I also want to run the tests on a few other makes /models as
    time allows.

    It might also be interesting to conduct similar tests on the
    same cam as it ages to see if performance degrades with age.

    Take Care,
    Dudley Hanks, Feb 6, 2014
  6. The native ISO of most sensors is slightly above 100, so
    it's very likely that ISO 200 would be the best minimum
    to use for testing.

    Except, given that you are looking at a JPEG, you'd no
    doubt be better off to crank up the ISO amplification a
    bit. Try it at ISO 800, where you should see
    significantly higher noise.

    If you can, the best way to do this (particularly across
    many different camera models) is to install DCRAW and
    use that as the RAW converter. That way you can get
    exactly the same processing for every different camera
    model. And use TIFF formatted output rather than JPEG.

    It still won't be good "science", but it can be fun and very
    A couple of quirks, in addition to the above strangeness at
    around the native ISO of the sensor... You might (if your
    technique is good) see odd variations at other than even ISO
    steps. 100, 200, 400, 800 and so on should produce a smooth
    curve; but increments of 1/3 or 1/2 stops between 100 and 200
    or between 200 and 400 will almost certainly not be smooth!

    On most, but not all, cameras there will be higher noise
    at ISO 126 than at 100, but less noise than either of
    those at ISO 159. Then ISO 200 will be just higher than
    ISO 200, as expected, but ISO 251 will be much higher
    and 318 will be lower.

    That is the effect of using a digital signal to control the
    gain of the analog amplifiers. The device doesn't actually
    increase gain, but instead switches in and out a lossy pad.
    It should be fairly stable. Less variation over say 5
    years than between your tests done 5 minutes apart! :)
    Floyd L. Davidson, Feb 7, 2014
  7. Dudley Hanks

    Dudley Hanks Guest

    I was aware of how JPEG would affect the results of the test,
    which is why I also used a PNG version as a more accurate
    representation of the true noise level. As noted in the
    original post, the PNG contained about half the noise of the

    The reason I included a JPEG in the test was to get an idea of
    how much that format can skew the results, so I can more
    accurately judge any test I might conduct on a cam that doesn'
    t offer a RAW, TIF or PNG (non-lossy) option.

    In particular, I'm interested in how my Canon SX120IS ranks
    alongside the XSi.

    Take Care,
    Dudley Hanks, Feb 7, 2014
  8. Dudley Hanks

    Dudley Hanks Guest

    I just ran the test on my SX120Is, and noted the following:

    * Fewer pixels registered values higher than 0 (about 2%)

    * The value of the noisey pixels tended to be higher (around
    80 on the 256 gray scale)

    It should be noted that there was no RAW file to examine, only
    a JPEG. So my pixel count was performed post processing ...

    It might be interesting to change to a different mode (
    landscape, night, etc) and see if the noise level is different.

    Take Care,
    Dudley Hanks, Feb 7, 2014
    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.