Originally Posted by Der_W
Just to make sure I'm getting the problem right: when you're doing a Gaussian blur based separation to fix problem areas in your image, once you work on areas of high contrast, you run into problems?
The first thing I'd try then would be to use something different from Gaussian blur to create the separation.
E.g. the "Filter"-"Noise"-"Median" preserves (especially round) edges pretty well. Or you could try the "Filter"-"Blur"-"Surface Blur". Possibilities are nearly endless :-)!
But if the asymmetric separation works, that's great as well, I just thought this could give you even more flexibility :-).
Yes, Jonas. I was finding that gaussian blur spread the color and luminosity too much on the low frequency layer in areas where there is a strong sudden shift in color and or luminosity. Imagine for example an image of a red apple set against a blue background. A normal, non-asymmetrical split is applied. If you clone on the high frequency layer in the area where the wall meets the apple, the cloning will not look good because both the wall and apple colors are contaminated by each other.
I will try your suggestions to use surface blur or median filter. Thanks.