4

Many articles mention color (chroma) noise. However, I cannot find proper information what is the cause of this kind of noise.

It looks quite different than the luminance noise, so for this reason I assume, the cause is different.

Some discussions mention a demosaicing algorithm, some mention the heating of the sensor.

Improve this question
2
  • 2
    Do you know how a digital camera produces coloured images at all, and in particular what a Bayer matrix is?
    – Philip Kendall ♦
    Jan 13 at 13:43
  • Yes, I do. But I am not sure if the color noise is just a luminance noise where the color is more visible for some reason (because every noise with color mask must be colored right?), or if it has a different cause.
    – Holiweil
    Jan 13 at 20:31

2 Answers 2

Reset to default
5

As the camera cannot really see color, there is an array of color filters in front of the sensor that detect basically only brightness (for that color due to the filter). The most relevant pattern for these color filters are called a Bayer matrix. Some camera manufacturers use a different pattern, namely e.g. Fuji with their X-trans Sensors.

Bayer Filter on Sensor

This means that the RAW file processing does something called demosaicing. Basically several hardware pixels or photo sites in the sensor are grouped to form an image pixel. In a Bayers matrix, this is one red, one blue & two green ones. The three colors red, green and blue (RGB) are then used to determine that pixels color.

If there is much noise in the image - which means randomized jitter of the correct brightness value, it can lead to the colors becoming pretty unstable. If the three color jitter values are e.g. less red, less blue, more green, then you will end up with a pretty pronounced green cast of that pixel's color.

So usually the brightness levels of color noise is less severe as it is evened out a bit by the grouping of the hardware pixels, but the color shift can be pretty severe.

So: The cause is really the same: Signal to noise ratio becoming worse. However the effects in the final image are different.

Improve this answer
8
  • 2
    If you think about it, this explanation (which I'm not disagreeing with, b.t.w.) highlights that there's really only luminance noise - it's just that, after being filtered, demosaiced, and whatever other processing steps are taken, some of it the noise is interpreted by our brains as color or chroma noise while other bits look more like luminance noise...
    – twalberg
    Jan 13 at 19:58
  • Thank you KaiMattern and @twalberg for your very detailed answers! It makes sense and it was one of my inner explanations. The problem was all these articles treated the chroma noise as something completely different. Maybe one more question: It seems, the chroma noise is more clustered (the colors stick together and create small blobs). Am I right or it is just my brain and illusion?
    – Holiweil
    Jan 13 at 20:35
  • 2
    @twalberg, that's not entirely true. Color noise is really a form of Moire pattern that occurs because demosaicing interpolates pixels of the three channels from slightly different places in the bayer filtered grid. Moire patterns are usually thought of in the misalignment of straits lines on a grid, but when applied to random fluctuations you get moire patterns on the pixel level that are just as random as the noise they represent. Think of it as a second order derivative that is applied on top of (not in place of) the first order luminance noise.
    – LightBender
    Jan 13 at 20:51
  • 1
    @Holiweil, they are actually very much the same thing, they differ only in their organization. Your first example is merely occurring on straight lines, so the artifacts form straight lines (long strings of pixels affected in a similar way), if you apply the same principle to random noise, you get random artifacts. (your second image has luminance noise, color noise, and jpeg artifacts form a couple different stages of the compression, making it even worse)
    – LightBender
    Jan 13 at 21:46
  • 1
    It's probably wise to acknowledge that Bayer masks are not the same "red", "green", and "blue" colors as the red, green, and blue pixels on our emissive displays are. In fact, the colors of the filters on Bayer masks are closer to the colors to which of the three types of retinal cones are most responsive. "Blue" filters in most Bayer masks are closer to blue-violet, "green" filters have a slight amount of yellow mixed with the green, and "red" filters are nowhere near red. Rather, they're somewhere between yellow-orange and orange. It's the "red" filters that tend to vary the most...
    – Michael C
    Jan 14 at 7:04
1

Noise in electronic circuits: Signals conveying information always contain static, random hissing on radio signals and random speckles on TV and digital imaging. As to noise in digital imaging, the signal (charge) generated by photon hits on the imaging sensor are incredibly weak. To be useful, these charges must be amplified and converted from a charge to a voltage. The sensor chip contains millions of tiny receptors, converters, and amplifiers. These should all be identical, however, each works at different efficiencies. Now each outputs a fraction of the image signal, however each signal contains unwanted static. This phenomenon is called the “signal to noise ratio”.

We see this static as random speckles that degrade our images. We have random noise seen when the amplificon is set high (high ISO). We have fixed pattern noise when the ISO is set low. We have banding noise induced by the camera software. We have Luma noise, which is monochromatic and Chroma noise, which is speckles of red, green, and blue.

Bottom line, the photo engineers still have work to do to make a camera that reproduces a faithful image.

Improve this answer
1
  • Thank you Alan for your detailed explanation. But the cause of luma noise and color noise is the same right?
    – Holiweil
    Jan 13 at 20:38

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.