Image deblurring by generalized total variation regularization and least squares fidelity

摘要

Motivated by the capability of ℓp-regularized algorithms and the close connection of total variation (TV) to the ℓ1 norm, a pth-power TV denoted as TVp is proposed with the condition 0 ≤ p ≤ 1. It's difficulty to handle the TVp-regularized problem for image deblurring on account of the nonconvex part in the objective function, then we tackle this problem by a weighted TV (WTV) minimization where the weights are updated iteratively to locally approximate the TVp-regularized issue. The complexity of WTV minimization is solved by a revised split Bregman approach. Numerical results are presented to demonstrate the improved performance of the proposed algorithm with p <; 1 relative to traditional deblurring methods on a variety of images.