Research Of Image Restoration Algorithm Based On Sample And Sparse Decomposition

With the development of image restoration in recent years,image restoration has influenced people's life more and more deeply because of its extensive practicality,so more and more scholars began to study the field and put forward a lot of algorithms.Image restoration is a technique to effectively estimate the damaged area by the intact information retained in the image and to spread the obtained data to the unknown information.The purpose of image restoration is to fill the damaged image or remove one of the objects in an image to achieve the desired effect.The quality of the repair depends on whether the observer is aware of the signs of repair.Image restoration techniques are widely used to repair damaged images,restore valuable artifacts,and remove target objects that people want to remove.Currently,the classic image restoration algorithms are divided into two categories:based on partial differential equation repair methods and image block-based image restoration algorithms.The former is suitable for repairing small structural damage images,and the latter repair method can take both structural and texture restoration into consideration.Based on the above two kinds of algorithms,this paper has made improvements to address their deficiencies.(1)A multi-sample block matching image restoration algorithm is proposed.The Criminisi image restoration algorithm for the image which has the complex texture structure information around damaged area,there is a case where the priority is unreasonably and the mismatch of the matching blocks.There for a multi-block matching method is proposed.To the priority,in order to make the structural information more repaired,the structural term is improved to the sum of the block structure sparseness and the neighborhood pixel difference degree.To the match block matching unreasonable situation,according to the two variables of the distance between the block and the distance inner the block,a threshold is set for the distance between the blocks,and a threshold is set for the distance inner the block.If the distance between the blocks is greater than y or the distance inner the block is larger than ?,the result will be deviate when the broken block is matched with the best sample block,so the missing information is synthesized by the weighted average of the sample blocks to repair the broken block,otherwise,using a sample block copy method.The results show that using this method repairs the results is better.(2)An image restoration algorithm based on edge extraction and sample combination is proposed.In the Criminisi image repair algorithm,due to the priority is not reasonable,it easily leads to the structure is not smooth,and when calculates the structure,it will be affected by the texture factors,resulting in the structure of the calculation is not accurate,therefore,the image is broken down into structural components and texture components.The edge of the damaged image is extracted on the structural component map,and the damaged edges are repaired by Criminisi algorithm to achieve the purpose of rough repair structure.Finally,the improved Criminisi algorithm is used to repair the structure component of the repaired edge is combined with the texture component.In the improved Criminisi algorithm,the priority structure term is calculated using the structural components,eliminating the effects of the texture.The data items of the priority function replace the original image with the repaired structure components,In order to reduce the error matching of the matching block.the inter-block distance and the intra-block distance are used to set the threshold value.In the case of false matching,a multi-matching block image restoration algorithm.On the contrary,using the Criminisi'algorithm repairs the image.Experimental results show that the restored image not only has more structural connectivity,but also achieves satisfactory results in texture restoration.