Improved algorithm for image TV regularization restoration model based on texture and contrast compensation

Abstract

Adopting discrepancy principle as iteration stopping criterion, Bregman iterative algorithm for image total variation (TV) regularization restoration model has attracted vast interests in the recent years. To a certain degree, Bregman iterative algorithm overcomes the shortcomings of TV regularization model: prone to reduce image contrast and prone to excessively smooth texture. However, some texture and contrast of image to be restored still exist in ultimate residual image. Based on analysis of nonlocal means (NLM) algorithm which is guided by a reference image, this article presents an improved algorithm, which extracts some texture and contrast from the residual image and then compensates them to the restored image of Bregman iterative algorithm. This improved algorithm can overcome the shortcomings of TV regularization model further. Numerical experiments show that the improved algorithm based on texture and contrast compensation can increase the quality of restored image.