Research On Image Inpainting Algorithm Based On Rough Data-Deduction

With the innovation of digital image processing technology,image restoration technology has become a new research hotspot.Digital image restoration technology is to analyze the relationship between the known pixel information and the missing area in the damaged area by some calculation method,and to repair the unknown area by some rules using the information reasoning of the known area,so as to achieve the purpose of structural integrity and visual connectivity.Image restoration technology is widely used in the restoration of old photos,the protection of antiquities,the restoration of works of art,the detection of public security cases and aerospace photography.In order to simplify the process of manual image restoration and improve the efficiency of manual image restoration,many image restoration algorithms have been proposed.In order to mine the potential and quasi existing relationship between image data,this thesis researches the rough data-deduction theory on the basis of image restoration algorithm,and focuses on the Criminisi algorithm and BSCB algorithm.Criminisi algorithm,as a representative of excellent image restoration algorithm,gets better visual effect when repairing some damaged images,but it also has its shortcomings.When the algorithm searches for matching blocks,the information provided by the blocks to be repaired is less,which results in a small matching range when searching for matching blocks.To solve this problem,this thesis proposes an improved Criminisi image restoration algorithm based on rough data-deduction theory.Rough data-deduction aims to expand search space,increase search data,expand search scope and deepen search depth.The algorithm proposed in this thesis has the following improvements in the search rules: the image content is divided into a data domain through the image structure information,and then the information amount of the block to be repaired is expanded through the rough data-deduction,and the search range of the matching block is expanded,so as to search the matching block and repair the damaged image.Experimental results and data show that compared with the classic Criminisi algorithm,the improved algorithm can expand the amount of matching block data,search more data,obtain better visual effect,and improve the peak signal-to-noise ratio of the image.BSCB algorithm,as a representative algorithm of image restoration in small damaged area,iteratively transmits the known information of the edge of damaged area to the damaged area to repair the image according to the partial differential equation and the thermal diffusion equation in physics.Based on the study of BSCB model,it is found that the Laplace operator introduced in the transmission process uses four neighboring points around a pixel,which will have limitations on the representation of pixels,and then cause the phenomenon of edge blur after repair.In order to optimize this problem,this thesis proposes an improved BSCB algorithm based on rough data-deduction,which uses the rough data-deduction space to formulate the rules associated with a certain pixel to mine the approximate relationship,derivative relationship and expansion relationship between pixels,and select the point with the greatest correlation with a certain pixel,so as to avoid the local problem of pixel representation.Experimental results and data show that,compared with the classical BSCB algorithm,the improved algorithm takes more points in the transmission process to reflect the image structure and obtain better visual effect.The peak signal-to-noise ratio also confirms the improvement of the restoration effect from the data level.