A New TGV-Gabor Model for Cartoon-Texture Image Decomposition

Abstract

Integrating the advantages of two recently developed total generalized variation (TGV) and Gabor wavelets, this letter presents a new weighted TGV-Gabor model for the challenging problem of cartoon-texture image decomposition. Computationally, by introducing two dual variables, we formulate a highly efficient numerical method based on the primal-dual framework in detail. At last, in comparison with several existing advanced variational models, experimental simulations clearly illustrate the outstanding performance of our proposed edge-preserving model, especially in separating the larger structural features from the smaller textural details completely and maintaining the sharp edges and weak contours simultaneously.