Research On Technology Of Image Inpainting Based On Sample Texture Synthesis

Nowadays, Digital image inpainting is one of the hot issues in image processing. It hasgreat theoretical value and application prospect for recovery of ancient artifacts, removalof text or obstacle in pictures, recovery of incomplete information, video error concealment,image scaling, ect. This paper has studied the digital image inpainting algorithm. The mainwork of the paper as follows:First, it studies on three kinds of variational PDE-based digital image inpaintingmodels: BSCB model, TV model and CDD model. It analyzes the performance of threemodels’ algorithm and limitation, which only suitable for fixing image line structure andcannot restore texture detail.Then, it focuses on Criminisi algorithm based on sample texture synthesis. Simulationanalysis discovers that Criminisi algorithm can combine structure inpainting and textureinpainting, cause the restoration complete in one process to ensure the continuity of imagestructure, maintain texture detail and integrity, and accelerate inpainting speed. Thealgorithm has a good inpainting effect on damage area with rich texture information andsimple structure information.Finally, it improves Criminisi algorithm based on sample texture synthesis. Whenusing global search pattern to search for optimum sample matching block, Criminisialgorithm is both time-consuming and prone to make error match. Because of thisdeficiency, FCM average clustering method is used to preprocess texture segmentation ofincomplete image, and limit the research of optimum sample matching block to a numberof texture segmentation areas associated with blocks which will be fixed, which not onlyimproves inpainting efficiency, but also obtains better inpainting effect. Determiningpriority function calculation of inpainting order through Criminisi algorithm is not reliable,which affects inpainting effect. Owing to this problem, increasing weight value isintroduced to enhance the role that structure information determines inpainting priority ofpixel pels of incomplete area boundary and to obtain better, more reasonable inpaintingorder as well as better inpainting effect.In addition, this paper performs simulation experiment for improving the algorithmand has confirmed its feasibility and validity.