0

Modern phones have some amazing processing of the data from their cameras. For example, the raw image below of a sculpture in a park after sunset (demosaiced using IGV method in RawTherapee and developed in the "Neutral" profile) has some colorful noise, which is totally colorless in the JPEG shot of the same scene with the same parameters (shots done using RAW+JPEG mode in OpenCamera). The photos were taken using Samsung Galaxy A320F/DS at aperture f/1.9, shutter 1/16s and ISO 2149. I've also seen similar processing in the Raspberry Pi camera V2.1 (Sony IMX219), and many of the modern DSLRs appear to also lack any colorful noise in their JPEGs, still having colorless noise.

I wonder, what algorithms are used to achieve such conversion of colorful noise to colorless one? Is it some high-ISO-optimized demosaicing? Or is it a special denoising algorithm applied after demosaicing? Or something else?

Raw:

demosaiced raw shot

JPEG:

corresponding JPEG photo

Improve this question
2

I wonder, what algorithms are used to achieve such conversion of colorful noise to colorless one? Is it some high-ISO-optimized demosaicing? Or is it a special denoising algorithm applied after demosaicing? Or something else?

Yes.

There are different algorithms that can be applied at various stages of processing. You can look at a raw processing application, such as Raw Therapee, to see what methods are available. Some demosaicing methods are designed to reduce color noise and artifacts. Other noise reduction occurs after demosaicing. What a particular camera or phone uses depends on what the developers decide to implement.

If you need to reduce chroma noise in an existing image, a simple way to do so is to blur only the chroma channels.

Improve this answer
  • Actually, it appears that RawTherapee even has dedicated settings for noise reduction separately in chrominance and luminance channels. – Ruslan Jun 4 '20 at 21:01
0

The most common form of noise is called "fixed-pattern". This is a static like function that occurs when the camera logic turns up the amplification applied to the imaging chip. Their is an amplifier in each light sensitive photosite. Each will have a slightly different efficiency. When the amplification is turned up too high, to gain sensitivity so the camera operates in feeble light, some photosites that should image as white, display some gray or false color. Some that should image as black, display as dark gray or false color. Software can be written to identity these noise induced defects. Thus a correction can be applied to mitigate.

Improve this answer
  • My question was more concrete. Of course software can be written: it must have been written to put in the camera firmwares. But what algorithms have the particular feature of turning colorful noise into colorless noise? (Is this question maybe off-topic on this SE site?) – Ruslan Jan 15 '20 at 20:21
  • Most of the noise in this image is not fixed pattern noise, it is Poisson distribution noise due to the random distribution of light. – Michael C Jan 17 '20 at 2:10
0

There are many algorithms, for instance the Selective Blur in Gimp:

The Selective Gaussian Blur plug-in doesn't act on all pixels: blur is applied only if the difference between its value and the value of the surrounding pixels is less than a defined Delta value. So, contrasts are preserved because difference is high on contrast limits.

Working on your raw image:

enter image description here

Improve this answer
  • This is an interesting approach, but this still leaves the lower-frequency chroma noise (particularly visible on the sky) that is totally absent in the camera JPEG. – Ruslan Jan 17 '20 at 21:13

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.