A NON-CONVEX NON-SMOOTH BI-LEVEL PARAMETER LEARNING FOR IMPULSE AND GAUSSIAN NOISE MIXTURE REMOVING

This paper introduce a novel optimization procedure to reduce mixture of Gaussian and impulse noise from images. This technique exploits a non-convex PDE-constrained characterized by a fractional-order operator. The used non-convex term facilitated the impulse component approximation controlled by a spatial parameter-y. A non-convex and non-smooth bi-level optimization framework with a modified projected gradient algorithm is then proposed in order to learn the parameter-y. Denoising tests confirm that the non-convex term and learned parameter-y lead in general to an improved reconstruction when compared to results of convex norm and manual parameter lambda choice.