Semiblind deconvolution

[reverofevil]

Algorithm could get some more data after blind deconvolution by user input. I.e. if one has a blurred image of a screenshot and deblurs it (with a lot of additional moire), now he has a chance to tell that some area has the same color, or that some area doesn't have a strong gradient in it. The idea is quite straightforward, so there should be an article on it somewhere.

(If you're not interested in searching for possibly non-existing article or implementing such algorithm, please, leave this feature request opened in case someone likes to try it and make a pull request.)