Non-convex variational model for image restoration under impulse noise

This paper proposes a novel non-convex regularization model for recovering the degraded images corrupted by impulse noise. The introduced model closely incorporates the advantages of total generalized variation and non-convex prior. This combination helps to eliminate the remaining staircase artifacts while preserving sharp edges and then results in the highly desirable restorations. To optimize the resulting minimizations, we develop in great detail two efficient numerical algorithms called primal-dual method and alternating minimization method. Several visual experiments and measurable comparisons, which indicate the visualization quality and restoration accuracy, are presented to demonstrate the outstanding performance of our new methods for image restoration under impulse noise over some well-developed numerical methods.