Making Foveon-like Images with a Bayer Camera

bizi clop wrote:

Tom Schum wrote:

Again, interesting.

Do you think LibRaw is working on the Bayer quads the way it should?

It seems to work properly.

But think about that a 2x2 square has 2 green pixels, which have a centre of "mass" at the exact center of the 2x2 square, but those lonely green and red pixels' centers are not aligned with the 2x2 square's center, thus causing a chromatic aberration-like effect (though it's not too serious, but it's still there).

LibRaw averages the two greens at the center. Yes the blue and red are slightly offset from this center, but the offset is half the diameter of the quad. So, it is far less than typical Bayer interpolation, in which blues in adjacent quads are interpolated into the center of a given blue sensel. Same for reds of course.

Ted's link shows it does not solve color moire though, and I guess you would need a separate diffuser over each quad, in order to eliminate the problem. This is not the same as a low pass antialias filter because it does not limit detail at the quad level.

Ted's link

Color photography without demosaicing in practice (kasson.com)

has the following, "The two images that were down-res’d with bilinear interpolation were both softer than their bicubic sharper counterparts, and didn’t have significantly fewer artifacts, so I ignored them. The differences between the images demosaiced with the two quality settings were minimal, so I ignored the bilinear interpolation demosaiced one.

Overall, the image processed without demosaicing had greater microcontrast, but more false-color artifacts and more evidence of luminance aliasing artifacts. This was a surprise to me."

What matters to me is microcontrast, and I get it without sharpening if I use superpixel deBayering. I like microcontrast, so it works for me. Not necessarily everyone of course.

Where to go from here?

If you look at the Bayer layout, and center a green in a 5-sensel area, that center green will be surrounded by two blues and two reds, with their centers in the center of your green sensel. Of course if you use this to make an image, each red and blue sensel is used twice and each green is used once. The output image would have half the pixels of standard Bayer. I don't think anyone has tried this, but what do I know. The overall resolution of this method would be not as fine as the quad by quad method, but it might be more resistant to color moire; I don't know. Might still need individual diffusers.

We assume each green sensel is covered by a filter with the same passband. In the future, it might turn out better to slightly change the passbands of each filter in the quad, in which case the two green filters would then be slightly different from each other, and green-centric decoding could not work.