My understanding of noise is that it's caused, effectively, because photos aren't evenly distributed accross time. So if you are shooting in low light, you get more noise because there aren't enought photons to 'average out'.

My question is this: I feel like there should be an algorithmic way by which you can sacrifice colour information (perhaps by going back to the demosaicing phase) of a RAW file to remove noise - effectively saying (well, the green pixels either side of this blue one didn't pick up as much light, so that's probably an error on this (white balanced) image).

Does such an algorithm exist?


Is there a way to use B&W conversion to reduce noise?

That depends on what you mean by 'noise'.

The conversion to B&W will effectively eliminate all chrominance noise.
It won't do much for luminance noise.

You must keep in mind that even though the values reported by each photosite (a/k/a pixel well, sensel, etc.) on a digital sensor are monochromatic, they're all filtered by one of three differently colored filters. If most "green" filtered pixels have a lower luminance value than adjacent "blue" filtered pixels, it most likely means that the light falling in that area has more "blue" than "green" in it. Noise reduction algorithms are more likely to interpret "green" filtered pixels that are brighter than other nearby "green" filtered pixels as noise.

The only real way to do what you suggest is to eliminate the Bayer mask altogether so that each photosite can be purely monochrome when the light is collected. There are a few monochrome digital cameras available that do just that.


Most noise is not caused by variance in the photon count. On an area as large as a sensor pixel (many times larger an a grain of high speed film) it makes very little difference. Instead it is literal signal noise within the electronics themselves which is then amplified along with the signal as the ISO is increased.

The algorithm you're describing is more or less how current noise reduction technology works, it uses context of the surrounding pixels to guess how much noise affects the current pixel. More advanced ones have edge detection and other features to improve the result. Even the, it's not very good.

When talking about black and white specifically, how you convert the image can greatly affect the amount of noise in the final image.

There are many methods for converting an image from color to black and white. The favorite (and Adobe recommended method) seems to be the "Black & White" adjustment. This method is actually not very good. It works by calculating the desaturated pixel and then multiplying the value based on its hue angle based on the sliders you select. This is essentially signal amplification, which also amplifies noise, so any slider with a positive ratio (a value above 50 in the Black & White adjustment) is also increasing the noise in those areas.

On the other hand, using the channel mixer uses a weighted average of three values. It is much easier with this method to avoid the signal amplification trap because all three channels sliders can be under 100%. The green and red channels usually have less noise than the blue channels, so you can lean on those two when possible to reduce it further. My go-to starting point is [R 60, G 90, B -40] then adjust from there.

Your Answer

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