Research On New Methods Of Image Inpainting Based On Criminisi

Digital image inpainting is a method that uses the known information of the image to fill the damaged area by mathematical methods and restores the image to the greatest extent.The application fields of image inpainting involve all aspects of people's lives,such as photo repair,removal of unwanted objects,and so on.So the research on digital image inpainting method is very meaningful.In this paper,the Criminisi image inpainting algorithm is studied in depth.The research status of image inpainting methods is summarized,the advantages and limitations of traditional image inpainting algorithms are compared and analyzed,new image inpainting methods emerging in recent years are discussed,and the evaluation indicators of image inpainting widely used by researchers are summarized.Based on the existing image inpainting methods,this paper has mainly completed the following three aspects:First.The partial differential equations based methods including BSCB(Bertalmio Sapiro Caselles Ballester)image inpainting model,TV(Total Variation)image inpainting model and CDD(Curvature Driven Diffusion)inpainting repair model,together with the texture synthesis image inpainting Criminisi algorithm were simulated respectively.The advantages and disadvantages of each model were summarized.The deficiencies of the Criminisi algorithm are proposed.Second.An image inpainting method based on sparse representation and texture decomposition is proposed.A texture term is added to the Criminisi algorithm to calculate the optimal weights to ensure that the filling order can be performed in the correct texture direction during image inpainting.Aiming at solving the shortcomings of the Criminisi algorithm in searching for the optimal matching block,a strategy for searching the global optimal matching block using the bus distance was proposed.Third.Because processing the original image directly in the airspace cannot balance the constraints between the various features of the image,this paper introduces a Nonsubsampled contourlet transform(NSCT)to decompose the image to obtain different feature sub-images.Each feature sub-image of the image is repaired,and then the Non-subsampled contourlet inverse transform is used on the restored feature sub-images to obtain the final repair result.Through simulation experiments,it is shown that algorithms proposed in this paper are superior to some existing image inpainting algorithms in terms of visual effects and objective evaluation indexes.Therefore,the proposed methods have certain practical application value and reference value.