Pushing the noise envelope

Steve Gillette, in a comment, provided a link about Kodak's new sensor filter array. dpreview are also carrying the story. Is this really true, or is something missing?


I did some thinking on this, and I'm not convinced they've got it nailed. The claim is that by using the luminance pixels noise is reduced. I'm sure it is - these pixels effectively capture 3 times the light of filtered ones. Then again, why didn't they go with an LAB type array, why use RGB as well as luminance?

Then there is the resolution thing. Without colour resolution, the luminance "resolution" is either good for B&W or is just noise in itself. (If one wanted B&W, then one would not need the filter array anyway.) With colour pixels now spaced at 4 sites, compared to 2 sites for a Bayer array, the resolution that can be offered is only going to be that of a sensor with a quarter the pixels. Any interpolation is just guessing, it doesn't provide true resolution. If there were only a quarter of the pixels, noise would be inherently lower as the sensors would be that much larger. I'm sure that the extra interpolation will lead to colour fringe effects. Physics getting in the way again.


Of course, Kodak also have some fancy software algorithms to go with the new array which I'm sure helps with noise reduction but that could well be the crux of this working. then again, Canon are doing a really good job with high ISO performance (in large sensors at least) and there must a lot of algorithms involved there. I'm sure I read somewhere that Canon are also increasing the sensor site size by reducing the size of all the other stuff on the chip: less gaps, bigger sites.