Blind Detection of Object Edges in Locally Blurred Images Based on Sparse Representation
In the face of complex images,the traditional fuzzy image detection and restoration algorithm has the problems of low accuracy of fuzzy edge detection and poor generalization ability,that is,it can not guarantee the infor-mation transmission effect after the restoration of fuzzy objects.In order to improve the performance of local edge de-tection for fuzzy patches,a blind edge detection algorithm based on sparse representation and structural similarity is proposed.Firstly,the image data is preprocessed by using the gray and normalization method to improve the computa-tional efficiency of the image,and then the optimal parameters of the H-La place distribution of the image are esti-mated by the Nelder-M optimization method to complete the edge detection and feature extraction of the fuzzy image block,and then the sparse coefficient is solved by the OMP algorithm to reconstruct the fuzzy image block.Finally,by using the method of down-sampling,the blurred image blocks are zoomed and combined in multi-scale,and the ima-ges reconstructed twice are fused to complete the restoration of local blurred objects.Subjective simulation results of blind detection and restoration of blurred images show that compared with other baseline algorithms,the EBD algorithm has higher brightness and clearer texture after detection and restoration.The objective analysis of the simu-lation results shows that compared with other baseline algorithms,the P index of EBD algorithm is improved by 34.60%,the E index is reduced by 32.92%,and the S index is increased by 3.40%,that is to say,the image restoration is more realistic and the fuzzy target detection is more accurate.To sum up,the blind edge detection algo-rithm of EBD blurred image target improves the detection power of blurred image blocks through image sparse repre-sentation,and effectively provides image resilience,which has important research value in the field of computer vision simulation.
万方数据
维普
