Image Inpainting Based On Improved Criminisi Algorithm

In order to ensure that the viewer is unable to detect the image to be inpainted during the process of viewing the image.Firstly,this paper analyzes the advantages and disadvantages of several classic image inpainting algorithms.Secondly,it is determined that the superiority of Criminisi algorithm in inpainting broken images.Finally,it is proposed that related optimization and improvement algorithms for the problems that the inpainting images cannot be reasonably optimized based on Criminisi algorithm.The main work and results of the thesis are as follows:1.In consideration of solving the problem that the priority of the damaged edge can't be effectively solved by the priority in Criminisi algorithm.Firstly this paper integrates the edge detection factor in the calculation of the data item to strengthen the resolution of the weak edge of the data item.Secondly,it is introduced the local feature detection item in the calculation of the priority.Feature detection items improve the ability of priority to discriminate the local features(straight line and curve features)of the damaged area image.Finally,for the broken blocks with different confidence value values,the segmentation adaptive algorithm is used to calculate the priority value,and the confidence item reducing quickly has impact on the calculation of priority issues.2.The block matching criterion for the Criminisi algorithm only adopts the color difference judgment factor,resulting in the inability to reasonably select the best sample block problem.Firstly,we use the matrix similarity and information entropy similarity algorithm to roughly select the set of 50 sample blocks that are similar to the repaired block.Secondly,combining the Euclidean distance algorithm containing color and local features with structural similarity algorithm to find the optimum sample block from the sample block set.Finally,after the optimum sample block fills the block to be inpainted,the confidence segment adaptive segmentation algorithm is used to solve the problem of rapid decrease of confidence.3.Different template windows are required to inpaint different images according to the Criminisi algorithm.Firstly,it is stored that traversing and calculating the entropy value of 11-by-11 image block with gray information(known pixels in the image block),and the center pixels of the image block is stored.Secondly,it determines the minimum information entropy value and the maximum information entropy value,and the mean value of minimum information entropy and the maximum information entropy as comparison thresholds;then storing the central pixel number of the upper comparison thresholds and the lower comparison threshold respectively.Finally,comparing the central pixels' number of the upper and lower comparing threshold to determine the appropriate template for image restoration.4.The paper addresses the question of the improved Criminisi algorithm costing too much time.Firstly,optimization algorithm for drosophila is introduced.Secondly,it is determined that the similarity between the optimization algorithm for Drosophila and the block matching algorithm of searching the best block.Finally,the optimization algorithm for drosophila is incorporated into the block matching of the improved Criminisi algorithm to improve image restoration efficiency.