Blind Restoration Of Single Motion Blur Image

Blind restoration of motion blurred images has always been the focus and difficulty in the field of image restoration.According to whether the blur kernel is known,it can be divided into blind restoration of blurred images and non-blind restoration of blurred images.This dissertation mainly studies the blind restoration technology of blurred images,analyzes some of the shortcomings of the existing blurred image restoration techniques,and proposes two improved algorithms:（1）Since in some images,the values of the dark channel are not clearly distributed on 0,and the values of the bright channel are not clearly distributed on 1,and the corresponding performance of the deblurring algorithm based on the two-channel contrast prior is not good,a method of combining the strong contour retention ability of the L₀ regularization strength prior and the sparsity of the dual-channel contrast prior is proposed to solve the problem.Under the maximum posterior framework,a blind restoration algorithm for blurred images combining dual-channel contrast prior and L₀regularization strength and gradient prior is presented,and an effective optimization algorithm is derived by half quadratic splitting.Finally,the real image is further approximated using the estimated blur kernel.The experimental results show that the proposed algorithm has certain competitiveness.（2）Inspired by the local maximum gradient prior,it is observed through experiments that the local minimum gradient value of the clear image will also be significantly smaller during the blurring process,and the local minimum gradient value of the blurred image is more distributed on 0.In order to reasonably utilize the local minimum gradient value,the local minimum gradient value is subtracted from 1,which is called the local minimum gradient prior,and L₁ regularization is used to constrain it.The difference between the local maximum gradient value and the local minimum gradient prior of the clear image and the blurred image is visually displayed from the statistical distribution histogram and the cumulative distribution diagram,and the proof is given from a mathematical point of view.In order to make full use of the local maximum gradient prior and the local minimum gradient prior,under the framework of maximum a posteriori,a blind restoration algorithm for blurred images based on extreme gradient prior is proposed,and a half quadratic splitting is used to deduce a Efficient optimization algorithm.Finally,the real image is further approximated using the estimated blur kernel.The experimental results show that the proposed algorithm has certain competitiveness.