Research On Improved Image Restoration Method Based On Criminisi And Block Structure Sparsity

Image inpainting is to fill the image missing information area,so that the inpainted image can reach or close to a certain visual quality.In the cultural relics protection,damaged image repair and recovery,target obstacle removal and high-resolution image reproduction,image inpainting has significant applications and research values.Criminisi inpainting algorithm and patch sparsity image inpainting algorithm are two typical image inpainting methods based on texture synthesis.Compared with the traditional variational PDE image inpainting technique,it has better repaired quality performance in the large damaged or rich texture area.However,the traditional Criminisi inpainting algorithm still has many shortcomings:?In the priority model,the confidence term decreases sharply after multiple iterations,but the data items remain steady,so the data items are subject to the confidence term which tends to be a zero value,and the priority is no longer reliable which resulting in the wrong fill sequence;?During the image restoration,the patch block information is not analyzed concretely,and the texture part is mistaken for the edge part and the texture extension;?Finding the optimal match block in the known image area(including the inpainted patch area)will bring high time complexity.And the problem with the based patch sparsity image inpainting algorithm is:?Image Edge and texture analysis based only on the gradient magnitude can cause edge detection errors;?The Euclidean distance of the image gray value information which is used to measure the similarity between the image blocks can easily lead to structural fracture within the image;?As with the traditional Criminisi inpainting algorithm,searching the optimal matching block globally also further increases the time complexity.To address problems existed in the conventional Criminisi inpainting algorithm such as the unreasonable priority and non-adaptive computing leading to error filling and texture extension,the global matches in update known regions not only increasing time complexity but also affecting final restoration quality while limiting search space by threshold decreaseing image inpainting quality.An improved Criminisi inpainting algorithm based on adaptive gradient classification and matching was proposed.Firstly pixels in image initial known regions were adaptively classified into three types such as smooth,texture and edge by gradient histogram;Secondly an adaptive patch classification priority function was used to overcome texture extension;finally the estimated patch type and adaptive size function by gradient were used to guarantee high matching efficiency and avoid update known areas by only matching proper patch type size in initial known regions.The experiments show the proposed method can avoid texture extension,decrease time complexity and improve image inpainting quality.To address problems existed in patch sparsity image inpainting algorithm,such as edge detection errors,image structure breakage and the global search bring the high time complexity.A sparse image inpainting algorithm based on Sobel and Laplacian operators for improved gradient priority model was proposed.Firstly two gradient graphs were obtained by the convolution of Sobel operator and Laplacian operator with the image,and the product gradient graph was constructed by point multiplication;Secondly the similarity criterion function of the sample blockswere constructed by the product gradient graph,the formula of the sparsity of the block structure were improved to improve the sparse degree value of the block structure and determine the regional feature of the sample blocks.Finally the size of the search area was determined according to the sparse degree value of the image sample structure blocks to improve the search efficiency.The experiments show the proposed algorithm can better maintain the coherence of the structural part and the consistency with the neighborhood information,improve the quality of the repair and reduce the time complexity.Compared with the proposed criminisi algorithm based on adaptive gradient classification and matching,it has a better repair quality.