Total Variation Blind Image Restoration

Blind image restoration is one of very challenging research topics in the field ofimage processing. It has great significance for improving the image quality andrestoring the image information. Regularization method is an effective way to solve theill-conditioned problem in such an inverse process of image restoration, and in whichthe total variation (TV) regularization method can effectively restore the image details.The paper focuses on total variation regularization theory and numerical minimizingmethods and proposed a new total variation blind restoration method exploring splitBregman iteration (TV-BDSB). The main research works carried out in this article areas follows:①Based on the survey of blind image restoration technology development status,the article made an in-depth study of basic theory of bind image restoration andmethods, focusing on the total variation regularization blind restoration method.②The process of blind image restoration is ill-conditioned, and the article studiedthe regularization method to solve such problem, made an in-depth analysis for theperformance of total variation, e.g. anisotropic diffusion and preserving image detail,studied the numerical minimizing methods of the total variation regularization model.③To solve the cost function using different mathematical strategy will lead todifferent performance of blind restoration algorithm. Base on the previous work, thearticle proposed a new total variation blind restoration method. The main idea isadopting the operator splitting and penalty technology, present a deformation of originalcost function by splitting the image for or the point spread function from L1norm and L2norm, then the article proposed and extended split Bregman iteration method to obtainthe converged solution.④In order to obtain the converged solution, the article studied the constraintabout image and point spread function during the iterative process. For the point spreadfunction, we proposed the dynamic threshold constraints to replace the non-negativeconstraints originally used in traditional total variation blind restoration methods.⑤The proposed total variation blind restoration method was tested on bothsynthetic and real-life degradations, comparisons are also conducted with some existingblind restoration methods recently published in the literature. The performance of thealgorithm was evaluated in terms of visual evaluation and objective evaluation. The proposed blind restoration method can effectively restore different blur cases,it runs orders of magnitude faster than traditional total variation blind restorationmethod. Compared to some resent blind restoration methods, the TV-BDSB method hasa great of advantages in terms of subjective evaluation and objective evaluation.