Block-based Image Restoration Algorithm Research

Image inpaiting technology is an important branch of image processing, and it is widely used in many fields, such as the protection of cultural relics, the production of film and television stunt, virtual reality, eliminating redundant objects and so on. It is currently a hot research topic in computer graphics and computer vision, and it is also in the field of application. The purpose of image restoration(inpainting) technology is to study and solve the problem of how to better achieve the damaged part of the image, and the image inpainting algorithm based on the effective information of the damaged image is auto repair. Many image inpainting method can simply repair to the damaged region, but it is easy to cause mismatch, speed slowly, the quality of image inpainting lowly and after the devices of repair cannot very well into the image itself. Therefore, this paper proposes a patch based image inpainting algorithm, this paper is combines the original patch sorting algorithm with the low rank decomposition, the variance sorting search algorithm and the fractal interpolation, which makes higher the image quality.The specific steps are as follows:firstly, the low rank approximation. Decompose the damaged image by low rank approximation, the repaired image matrix decomposes the low rank matrix and the sparse matrix, and singular value decomposition is carried out for low rank matrix, its aim is to find the best low rank decomposition, the results and sample error reaches the minimum in the Euclidean distance. Second, a patch based sorting algorithm. Construct the permutation matrix, and the matrix is processes one-dimensional transform, and for minimizing the processing results to get the shortest path, and the square area of the around of the each patch is searches the optimal matching patches, so as to obtain the collating sequence, and to estimate the final image by permuting different weighted average. Again, the variance sort search. The after processing image divides sequence patch and main patch, the current minimum aberration variance information between the sequence patch and main patch, it is greatly reduces the searching and matches the main patches and the corresponding encoding time for each sequence patch, at the same time, it can exclude a considerable part of the main patch search while maintaining the same image quality. Finally, fractal interpolation. To create a smooth and natural image is very important. However, after the variance sort search, the inpainting domain is still not up to the desired level, and it also cannot reflect local characteristics between two known adjacent. Therefore, it is necessary to improve by the fractal interpolation, it can improve the smoothness of the image inpainting and to get the higher accuracy.This paper is goes the full theoretical analysis, a lot of experiments are carried out on the proposed algorithm. Experimental results show this paper can obtain higher efficiency and at the same time also can keep in good repair quality, with low complexity, computational speed, and higher repair efficiency and so on, thus validates the feasibility and effectiveness of the algorithm.