Electrical Flow Algorithms for Total Variation Minimization

摘要: The total variation (TV) minimization framework is a very popular method for wide variety of image restoration problems. This comes in two variants: anisotropic, where the “smoothness” denoised measured by L1-difference neighboring pixel; and isotropic, measure based on computing localized L2-differences thus rotationally invariant. There was lot work obtaining efficient algorithms computingTV denoising. Most this effort focused anisotropic variant as it possible to exploit its connection maximum flow problem. In case isotropic variant, no longer holds context rely convex programming techniques, which results much slower running time. paper we develop an approach TV that electrical flows builds upon introduced [CKM + 11]. encompasses natural way both variants obtains times versions are essentially same. On with n pixels m relations, our algorithm produces solution that’s within 1 ǫ optimum time ˜