A novel global method for edge extraction under Poisson noise: game theory

Edge extraction is a central problem in image processing and it is a necessary step for computer vision tasks. In this paper, a novel global method P-GSG for edge extraction of image under Poisson noise is given, which is based on sparse representation. Furthermore, a game model which combines P-GSG with total variation denoising is proposed to get better results. As two players, P-GSG model can apply with iteration latent clean image to robustly get the gradient under the Poisson noise, on the other hand, TV denoising can get an edge-preserving latent clean image, which overcomes the shortcoming of over-smoothing. By cooperation and competition between two tasks, we can attain a satisfactory solution for this game model-Nash equilibrium. The algorithms of P-GSG and TV denoising are given. Based on above algorithms, it is obvious that alternate iteration method is easily used to solve this game model. The effectiveness of these two models is shown by numerical experiments.