Reducing chroma edge bleeding

[tomczak]

Chroma bleeding sometimes occurs at saturated red or (less often) blue edges, usually bordering dark green background (such as foliage). It seems to be especially visible in smaller-sensor cameras. The reasons are unclear but conceivably the phenomenon could be caused by 1) large red Airy's disks due to greater diffraction of longer wavelenghts (but why blue then?) and 2) over-smoothing the R and B channels during the demosaicking interpolation as speculated here:

http://www.dl-c.com/board/viewtopic.php ... roma#p3440

Den and I have tried the following technique and we both find it quite effective in reducing the defect. The idea is based on different edge contrast in the bleeding channel (such as R) and the green channel (which is a proxy of luminance): the R channel is softer at the edge than G.

Simplified Recipe (use R channel for the most common red bleeding; use B for blue):

1) Extract R and G channels from the image
2) Blur/High Pass R and G channels; use radius similar to the bleeding extent.
3) Special Trans./Difference of high passes: R minus G
4) Use the last output (as is or after brightening it) as a mask in Saturation transform (in HSL mode) to reduce the edge bleeding appearance.

Bleeding Chroma Edge: original

Bleeding_original.jpg (9.38 KiB) Viewed 3957 times

Bleeding Chroma Edge: reduced

Bleeding_reduced.jpg (9.41 KiB) Viewed 3955 times