Variational Image Decomposition Model OSV With General Diffusion Regularization

Image decomposition technology is a very useful tool for image analysis. Images contain structural component and textural component which can be decomposed by variational methods such as VO (VeseOsher) and OSV (Osher-Sole-Vese) models. OSV model is a powerful tool for image decomposition but the minimization is a hard problem because of solving the 4 order partial differential equations with complex finite difference scheme for Laplacian of curvature. In this paper we proposed an improved OSV model with general diffusion regularization. The general diffusion terms can be TV(Total Variation), Nonlinear diffusion(Perona and Malik) and Charbonnier regulerizers. Additionally, we also use L1 norm as data term inspired by TV-L1 method. We also use Split Bregman method for the easy implementation of the improved OSV model. Experiments show the proposed method is a valid method for image decomposition.