Regularized Deconvolution Algorithm Based On LO Norm And Total Vairation

Image deconvolution is an important branch of image processing, which has been usedin many fields. Image deconvolution has been used to restore a blurred image with noiseto get a similar estimate about real image. This paper has put forward a new regularizeddeconvolution model according to the character of the norm of L0and the total variation.The original optimization model has been divided into two optimization sub-models withseparation of variables. The preliminary experimental results show the efectiveness of thealgorithm.The observed result will be blurred duo to the characteristic of the instrument and thehuman factor when we collect image, for instance we forget to focus the camera or shakewhen we take photos. These demonstrate image deconvolution has the important theorysignificance and the practical application value. In the process of obtaining, transmissionand receiving, image will inevitably be mixed with noise. We handle the problem with anexisting method of regularization because of the ill-posedness of the deconvolution problem.The method of total variation(T V) is a popular method in image deconvolution, and ithas been heavily used in the model of deconvolution due to its capability of protecting imageedge and details. The L0norm of image gradient could approximate prominent structure, sowe introduced the regularization about the L0norm. By the definition we can see that the L0norm of X is the number of nonzero. With that being stated combining the norm of L0and the total variation to restore image is feasible and rational, so we can get the new model inour paper.It is now essential that how to solve the model. Because of involving a discrete metric,traditional gradient decent or other discrete optimization methods are not usable, so we useseparation of variables with introducing two auxiliary variables to solve it by alternatingoptimization method. The basic idea is decomposing a complex optimum problem into twooptimum sub-problems. And then solve them. So we can quickly get a similar estimateabout real image.