Research On Image Blind Restoration Algorithm Based On Approximate Greatest Common Divisor

Image blind restoration is an important research direction in image processing,which has great scientific research value and engineering application value.In this thesis,we study the image blind restoration algorithm based on approximate greatest common divisors(GCD),but the algorithm still has some shortcomings in stability and real-time performance.In view of the above problems,the regularized approximate GCD blind restoration algorithm for multiple images and the improved approximate GCD blind restoration algorithm for single image are proposed respectively to improve the restoration effect of the degraded images.First of all,the traditional algorithms have the problems of slow convergence and non-unique solution in estimating the original image and point spread function.In order to restore images in the absence of prior information stably,the Bezout-type and Sylvester-type approximate GCD image blind restoration algorithms are studied in this thesis.The experimental results show that the two algorithms have poor anti-noise performance,and the Bezout-type blind restoration algorithm has higher real-time performance than the Sylvester-type blind restoration algorithm.Secondly,in view of the Bezout-type and Sylvester-type approximate GCD image blind restoration algorithms are sensitive to noise interference,this thesis constructs a regularized constraint model,and proposes a regularized approximate GCD multiple images blind restoration algorithm.The algorithm uses the Bezout-type approximate GCD algorithm to calculate the point spread function,and then uses the regularized constraint model to perform non-blind deconvolution,so as to obtain the restored image.Through experimental simulation analysis and comparison,the improved algorithm has better anti-noise performance,and the image quality evaluation indicators are improved.The restored image maintains better texture details of the image.Finally,the traditional approximate GCD blind restoration algorithms require two degraded images,but some scenes do not satisfy this condition,so it is necessary to use the approximate GCD algorithm to restore the clear image from a single image.Aiming at the problem of poor real-time performance and instability of the approximate GCD blind restoration algorithm for the single image,this thesis further proposes an improved blind restoration algorithm for the single degraded image.The size of the point spread function is calculated by QR decomposition,and the point spread function is further calculated by the Bezout-type approximate GCD optimization algorithm,and then the restored image is obtained by the Wiener filtering deconvolution operation.Through experimental verification and analysis,the stability and real-time performance of the improved algorithm are better than that of the comparison algorithms.