Color-Camera Images: Comparing CFA Split and Debayering

The raw images captured by a color camera (OSC) carry a CFA (color filter array), and to turn them into usable color data there are two main roads: one is “CFA split,” the other is “debayering.” What effect do these two approaches have on subsequent detail and noise? I’d wanted to run this experiment for a long time, and this time I finally got to it.

Four Processing Paths

I set up four processing pipelines to compare:

1. Do CFA split first, then star removal, then up sample (2x).
2. Do CFA split first, then up sample (2x), then star removal.
3. Do Debayer first (RGB combined), then extract R, then star removal.
4. Do Debayer first (RGB separated), then remove stars from R.

Observed Results

• The images that did CFA split first are blurrier, but path 2 has more detail than path 1.
• The images that did Debayer first are sharper, but the areas where the background SNR is low show a lot of black spots.
• Path 3 and path 4 are exactly the same.

My Current Conclusions

• If you’re going to do CFA split first, then it’s best to do a 2x drizzle afterward.
• If you do Debayer first, then you have to deal with the background noise problem.
• One more conclusion: if you do CFA split first and use this image only for the RGB color, with the luminance channel shot separately, that result should be better.

But I should add one caveat: the image used in this test wasn’t calibrated (I just grabbed a photo on the spur of the moment), so the results may have been affected by that; a more rigorous approach would be to calibrate first and then test.