Image Restoration by Adaptive Neural Network Based on Co-occurrence Matrix Analysis

An adaptive Hopfield neural network algorithm based on co-occurrence matrix analysis for image restoration is proposed in the paper.By the algorithm,the co-occurrence matrix of each image region is calculated,with the texture feature extracted and the nonzero elements in the co-occurrence matrix clustered.A new concept,detail intensity,and its computational method are proposed in the algorithm.Detail intensity distinguishes accurately the flat from detail regions in images,and adjusts Hopfield network weight coefficients adaptively through a nonlinear function,such that the weight coefficients are suitable for image texture features.The iterative process of image restoration and the updating process of weight coefficients of neural network are executed alternately.The algorithm can remove noise in smooth regions and reveal details in detail regions,in conforming to human’s perceptual criteria.Comparative experimental results show that the restored images produced by the proposed algorithm have higher SNR and better vision effect than those by some conventional image restoration methods.