The Bayer sensor vs the Foveon sensor (part 2 of 2)

Crop of the bayerized image

This image shows a portion of the black and white bayer data, scaled 200% with "nearest neighbor". A full-color input image is transformed into a bayer greyscale image by topgmraw.py, discarding 2/3 of all the data in the original. A modified version of the dcraw converter is used. The modification is simply a loader for ".pgm" files as bayer sensor data---the core algorithms used by dcraw to perform interpolation are unchanged.

Crops from the two images

Here are some crops from the simulated bayer and full-color sensor images. To make crop about the same size on the screen, the bayer image has been scaled 200% with "nearest neighbor", and the full-color image has been scaled 300%, also with "nearest neighbor"

The images speak nearly for themselves: my prediction was that the simulated bayer image would have higher resolution, and that appears to be the case. However, I also expected that some degree of color fringing would be present. In fact, it was quite easy to find a blur value (simulated in-camera low pass filter) to apply before converting to a bayer greyscale file that gave quite satisfactory results.

While these results are unlikely to end the debate among digital photographers, they illustrate to my satisfaction that comparing the number of photosites does not give an accurate impression of the relative resolution of a bayer and a full-color sensor. I suggest that the distance between the centers of green photosites may be a more useful benchmark. If this is the case, a Foveon sensor must have about 1.5x as many photosites as a Bayer sensor to convey the same amount of edge detail.

Files currently attached to this page:

bayer-full.png	1.7MB
bayer-reconstructed-full.png	3.4MB
bayer.png	40.6kB
cactus-medium.jpg	517.5kB
cactus-small.jpg	59.6kB
cactus.jpg	679.9kB
comparison.png	113.1kB
dcraw.c	174.1kB
foveon-full.png	1.4MB
povtest.sh	307 bytes
topgmraw.py	480 bytes