A GNC Algorithm for Deblurring Images with Interacting Discontinuities
Antonio Boccuto, Monia Discepoli, Ivan Gerace, R. Pandolfi, Patrizia Pucci · 2002
In this paper we present a Graduated Non-Convexity (GNC) algorithm for reconstructing images. We assume that the data images are blurred and corrupted by white Gaussian noise. Geometric features of discontinuities are introduced in the model and the problem is formulated as the minimization of a non-convex function. We give a convex approximation of such a function and a family of approximating functions. Moreover, to analize the convex approximation, we prove an alternative duality theorem to implicitly treat discontinuities of images. The experimental results are more satisfactory than those obtained by some standard algorithms.