Wondering about the possible merits of a hexachromatic CFA, one that sample more points along the visible light spectrum rather than the typical 3 RGB.
Is there any reason why such CFA scheme has not been tried?
Some early digital cameras were CMY; of course, most color printers always include cyan, magenta, and yellow inks as a minimum.
However, CMY are the subtractive primaries, useful when you start with white paper as your brightest tones possible, and the inks subtract from this white. Any other primaries will lead to dark prints. The brightest, most saturated blue pigment is rather dark, and compromises have to be made.
RGB are the additive primaries, when we are directly dealing with light.
The benefits to that approach would be.
- Higher Color Accuracy, specially skin tones. RMY
As C=G+B; M=R+B; and Y=R+G, I see no additional benefit from having three more filters, as they will duplicate what's already captured. Furthermore, the Bayer array of 2x2 color filters already lead to color error, which would be made worse by having more colors, and sensor resolution would decrease.
The human eye only uses three color receptors, and so that would seem to be sufficient.
Probably lower dynamic range, as those mixed colors would involve additional subtractions of one color channel from another to get an sRGB value out of the raw data. A while back I calculated that a pure CMY sensor would have about an 18% lower signal-to-noise ratio compared to an RGB sensor, and if you include the read noise, the results would be even worse. A mixed RGB-CMY sensor wouldn't be quite as bad, but still not as good as RGB.
- Higher Sensitivity due to CMY
That's true. You'd get a higher native ISO rating.
However, it seems that chroma noise is much more objectionable than luma noise, and this will increase the chroma noise.
- Less noticeable moire artifacts given the wider spread and more complex pattern.
The color defects would become more complex and intractable, as we see on the Fuji X-Trans array.
- Probably more resilient to IR contamination, IOW better Reds with the help of MY
It depends almost entirely on the IR cut filter, which can be used with any filter array.
In CMY, R = (Y + M - C) / 2, which would have more relative noise than just an R by itself.
A bigger problem with any color filtering is that we don't have a good match with human vision. Modern filters are close, but aren't quite there yet.
The only disadvantage I see is more complex processing algorithms but given today processing power that shouldn't be an issue.
That wouldn't be much additional processing, so you are right, it wouldn't be much of an issue.
