Karim D. Ghantous
Well-known member
The Bayer filter, in principle, is a terrific idea. A hidden feature is that it acts as data compression... of a kind. The problem with almost all Bayer filters is that there is a green bias. That's a huge problem that I'm surprised still exists.
In fact, even if all three primaries were represented equally, you would still have a problem with the blue channel, as it is the least sensitive. But blue is represented by only 25% of the pixels. Hence, blue channel noise.
I think I have found a method, in theory, that can compensate for this. It will mean an effective lowering of ISO, but modern sensors can handle that. Basically, what you do is use a filter that transforms the bias away from green and towards blue. So, perhaps a purple filter that lets through the following levels for each primary:
20% Green
40% Blue
40% Red
Your ISO will be reduced by at least a stop. Which is not a big deal. When you perform a white balance, the noise in each channel should be about the same, given a white light source.
I don't have the material do test this idea, but if anyone does, I'd love to know if this actually works.
In fact, even if all three primaries were represented equally, you would still have a problem with the blue channel, as it is the least sensitive. But blue is represented by only 25% of the pixels. Hence, blue channel noise.
I think I have found a method, in theory, that can compensate for this. It will mean an effective lowering of ISO, but modern sensors can handle that. Basically, what you do is use a filter that transforms the bias away from green and towards blue. So, perhaps a purple filter that lets through the following levels for each primary:
20% Green
40% Blue
40% Red
Your ISO will be reduced by at least a stop. Which is not a big deal. When you perform a white balance, the noise in each channel should be about the same, given a white light source.
I don't have the material do test this idea, but if anyone does, I'd love to know if this actually works.