Research On Image Restoration Algorithms Based On Bayesian Theory

Image restoration is one of the core issues in image processing field. It aims to make the blurred image more clearly by designing a mathematical model and its algorithm based on the information of actual image degradation. However, estimating the point spread kernel function and building the mathematical model according to the process of image degradation as well as priori knowledge about the image have always been the key problems for researchers to solve. Then we summarize the basic theoretical knowledge of image restoration, quality evaluation methods of restored images, and the adoption of the latest research progresses for Bayesian methods in the process of image restoration. In this thesis, the following problems about image restoration are studied:Firstly, the basic mathematical models for image restoration are elaborated, including the general degradation model and the priori model of image restoration. So we can conclude that our study of image restoration focuses mainly on two key issues:(1)The degree of mastering the priori knowledge of image restoration.(2) Which method can be adopted to the process of image anti-degradation, in order to obtain the original high-quality image.Secondly, this thesis manly focuses on the study of the theory of image restoration, which is based on the Bayesian theory. The maximum likelihood estimation, maximum a posterior estimation and Bayesian estimation are introduced respectively. And we focus on the improved Bayesian estimation process in detail. Finally, we propose a method that combined the sub-regional detection and Fuzzy filter to eliminate ringing effect.Finally, some common Bayesian parameter estimation algorithms used in image restoration are studied. We propose the algorithm of image restoration based on Bayesian theory mathematically, which starts with the general theory of parameter estimation and a priori distribution of the image, combined with the classical parameter estimation problems in image restoration field. And then we have done some mathematical experiments by using the empirical analysis method.As the process of image restoration is highly sophisticated and actually there are various reasons for image degradation, this thesis has a lot of room for improvement. For the commonly encountered problems in the algorithm of image restoration based on the Bayesian theory, more work can be done in the aspect of simplifying parameter estimation in the future.