willie_901 said:
Conor wrote:
"but then I realized - the color filter array being what it is, you'd have to interpolate each photosite with its nearest neighbors to get a pixels that looked panchromatic, instead of like a checkerboard."
Using information about nearest neighbor pixels is used in medical image analysis software (MRI for instance). so, don't let this stop you.
willie
The idea of interpolating between pixels isn't what stopped me -- what stopped me is the realization that this is precisely what a raw file converter already does.
🙂 So, I doubt I'd be able to get much of an image quality gain unless the bayer filter itself was removed as well.
Or, at least, this is my line of thinking. When uncompressed, the sensor data in the raw file probably looks like a long string of 12-bit integers (probably padded out to 16) interleaved like:
green,
red,
green,
blue,
green,
red,
green,
blue...
If you were to put all those "sub-pixels" into a bitmap instead of interpolating between them to make a full color image, you'd wind up with the very high-resolution monochrome image you'd wanted, but it would have a checkerboard pattern, because of the sharply different spectral response of each photosite in the sensor array.
I don't know how canon's pro raw-file converter makes higher quality b&w images than other competing converters, but I doubt it can somehow "invent" the missing 2 of 3 color channels at each photo-site. On the other hand, if the engineers know the precise spectral response of the bayer filter (which they probably do) and if they also know the color response characteristics of, say, tri-x, (which they also probably do) then it should be possible to create a grayscale image that matches the "look" of tri-x, with sensor noise instead of grain. Or t-max. Or whatever continuous tone, visible spectrum film they want.
To digress (and even to get back on topic), even with a spectral response that precisely matches fp4, i doubt most modern sensor cameras have enough exposure latitude to make its deep shadows look convincing (my normal development routine for fp4 yields I think 12 stops of exposure, I've measured it with a densitometer), and I'm skeptical of grain-generating algorithms. At this point, I'd rather just shoot the real thing for a buck-fifty a sheet and let the digital technology mature.
