Aapplications Of The Split Bregman Method In Image Processing

In the early 1970s, image processing has already developed relatively complete system of academic disciplines. It mainly includes:image segmentation, image denoising, image registration, image fusion, target detection, image compression, image reconstruction and pattern recognition, etc. Image processing have been successfully applied to the military, criminal investigation, business, medicine, art, home entertainment and other fields, and the application area is still expanding.This paper reviews the active contour model based the variation calculus and level set, and detailed the parametric active contour model, curve evolution theory, the level set method and the Chan-Vese model. Based on this work, many scholars made a variety of improvements from different the angles, but because of the drawback of the model itself or computing approach, there are shortages of their own, in which computing speed is a common problem; In addition, this paper expounds the four face recognization methods via the subspace-based, PCA, LDA, MMC, MFA. Compared with SRC method which we will improve, These four methods have the defects of low recognization rate and poor computing speed to varying degrees.This paper describes the classical Bregman iteration method commonly used for solving the convex functional extremum problem, and apply it to solve the constrained optimization problems with detailed steps. With a deep understanding of its principles, We show the Split Bregman method proposed by Tom Goldstein in detail, applied it to the process of solving the classical Chan-Vese model and got good results, greatly improved the segmentation efficiency. Since then, many scholars paid great attention to this method, and successfully extend its applications to ROF denoising, weighted GCS model, CCD image restoration model and other ones. This paper utilizes the Split Bregman method to the face recognition for the first time, to solve the SRC model. Based on a large number of numerical experiments, it shows that the recognition efficiency is remarkably improved with a similar or even higher recognition rate, verifies this method to be a efficient tool for solving large scale problems similar to the L1 regularizations problems, and further expand the scope of application of the method.