1

I'm trying to understand what is causing chroma noise in an image sensor, specifically why it is so visible when the ISO is set to a very high value (which for CMOS I guess means that there is an amplifier with a high gain value before the readout ADC?). Its noise structure appears to have a very low spatial frequency behaviour since I need a digital image processing filter with a very large kernel to remove it. I don't think this is being caused by typical photon shot noise. Perhaps some form of dark current shot noise or possibly thermal effects?

Any comments, suggestions or links to papers etc would be most welcome. Many thanks!

  • 1
    Maybe electronics.SE is a better match for this question? – inkista May 21 '15 at 21:46
  • I'm happy for this question to be moved. My reasoning for posting the question here is that there seems to be very knowledge sensor folk answering questions here – trican May 25 '15 at 20:10
  • 2
    Physically there is no such thing as "chroma noise" because a pixel (sensel) does not even know what color will be produced from its output. There is just "regular" noise and image processing. High noise level makes the demosaicing algorithm fail and create color blobs. Depending on the settings, you can get more or less color noise or more or less detail from the same raw image. – szulat Jun 21 '15 at 18:16
  • 1
    @szulat, for the most part I'd agree with that statement, but depending on the CFA position and by association the wavelength of the photon, the photon will penetrate the silicon deeper for one colour component compared to another. I disagree that the chroma noise is due to demosaic failures – trican Jul 13 '15 at 8:03
1

Shot noise goes with the square root of the number of photons, so every time you have 1/4 the photons you have 2x the noise (with respect to the signal). (Example - 400 photons has 20 average noise, 100 photons has 10, so go from noise is 1/20 to 1/10.)

Note that pixel electron capacity goes down as ISO goes up, as there is a limit as to how far you can multiply the signal before you run out of range. For example a Nikon D800 has 81,000 electron capacity per pixel at ISO 100 but only 175 at ISO 51,200. White is still a full pixel, so you can see how much more noise you will get by applying the first para.

Also the read noise (from sensor/electronics) gets amplified by the ISO amplifier along with the signal. If you have 10,000 electrons of signal 2 electrons of read noise is very little (especially compared to the average 100 electrons of shot noise). If you have 16 electrons then the 4 electrons of shot noise and 2 of read noise being to look very large, especially after the ISO amplifier multiplies the whole lot by 32x, or whatever.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.