Research On Exemplar-Based Image Inpainting Algorithm

With the rapid development of computer science and the advent of the digtal age,digital images have become one of the most important sources of information in people's daily life.However,there is a certain probability of loss of information during the formation,storage,and transmission of digital images,resulting in damage to digital images.Image inpainting technology is a method of recovering the lost information.It is used to fill the damaged part of the image or to remove unwanted obejects in the image.It does not leave any traces of artificial restoration,allowing neutral observers to think that the result of the inpainting is the original image.The digital image inpainting technology can be divided into a structure-based inpainting method and a exemplar-based inpainting method according to different methods for establishing a inpainting model.With the increasing demand for image inpainting applications,the field of image inpainting has been very active in recent years,and many excellent inpainting models have been born.The three most representative models are BSCB model,TV inpainting model and Criminisi algorithm.In this paper,the three classical models are studied in depth.The inpainted results of the three algorithms are compared through the simulation experiments.And some improvements based on the exemplar-based algorithm are proposed,including:(1)By analyzing the insufficient filling order of the original Criminisi algorithm,we define the pixel inhomogeneity factor to describe the local structure information of the image,instead of the data item in the original Criminisi algorithm.Aiming at the shortcoming that the confidence term of the original Criminisi algorithm approaches to zero rapidly in the iteration process,the form of the confidence term and priority function are adjusted reasonably.The improved Criminisi algorithm proposed in this paper is compared with the original Criminisi algorithm and the state-of-the-art algorithm.It proves that the improved Criminisi algorithm has the advantage in the inpainting of large damaged area image.(2)In order to overcome the defect that the exemplar-based inpainting method cannot flexibly change the size of the exemplar in the program process,we decompose the damaged image by quadtree in each iteration of the improved Criminisi algorithm.In this way,the improved Criminisi algorithm can adaptively select the best exemplar size according to the different local structure information of the area to be inpainted.Experiments show that the improved Criminisi algorithm with adaptive exemplar size selection has a good inpainting effect on the damaged images with complex structure.