Adaptive regularization for color image restoration using discrepancy principle

In this paper, we consider and study how to automatically select the regularization parameter in a color total variation minimization model for image restoration. The idea is based on that the variance of the noise can be estimated easily, thus the restored image should satisfy the Morozov discrepancy principle. We developed an iterative scheme to solve the color total variation (CTV) minimization problem, where the CTV norm is represented by the dual formulation and the proximal point method was applied to find a solution. During the iteration, the regularization parameter is automatically adjusted to guarantee the restored image satisfying the discrepancy principle. Numerical experiments are reported in the paper.