Research On Digital Image Inpainting Algorithms

With the development of information,digital image inpainting technology has been widely used in various fields as a digital image processing hotspot.This technology is a process of gradually filling the damaged area in the image,which makes the restored results satisfy people's sense of vision.It has been popularized in the fields of photography,medical treatment,cultural relic protection and so on.The image inpainting field mainly involves two directions based on texture synthesis and non-texture synthesis.This paper introduces the principles of the two types of algorithms in detail.Through the theoretical derivation deduction and verification of experimental data,and with the aim of improving the repair efficiency and reducing the time-consuming of the algorithm,the classical algorithms in each direction are proposed for improvement.The main works of this paper are as follows:The non-texture synthesis direction focuses on the CDD model algorithm,and proposes an improved algorithm for a series of problems such as iterative complexity and edge repair efficiency of the model.Considering the complexity of the iterative process,an optimized curvature term is designed.it can simplify the high order partial differential equation to reduce the running time of the algorithm;Then,introducing the ratio of pixel gray difference values to describe the degree of gradient change which makes the diffusion still along the direction of curvature and gradient;Finally,a weighted smoothing function was designed to improve the original diffusion coefficient,it avoid the phenomenon of mutual diffusion at the edge of the image.Experiments on image inpainting such as connectivity,fine scratches,and text breakage show that the algorithm effectively solves the problem of blurring and long repair time caused by CDD model.In the direction of texture synthesis,this paper focuses on the improvement of the formula of priority of Criminisi algorithm,the size of the sample block,and the search of the best sample block.Firstly,based on the texture features of LBP and gradient operator,the priority is reformed,which fully considers the texture features in the sample block to make the repair order more reliable.Secondly,After determining the maximum priority,the size of the sample block is adaptively selected according to the gradient and the degree of LBP change,and then the block effect and the error derived phenomenon at the edge are relieved.Finally,Using a combination of global search and local search to find for the best sample block,adaptively switch the search mode by setting the priority threshold,This paper uses the global search algorithm for Grey Wolf optimizer,save time and cost.In the local search,the distance item factors of the best candidate block and repair block are added in the original similarity measure,which reduces the probability of multiple candidate blocks appearing at the same time,and improves the accuracy of sample matching.Through image inpainting experiments such as object removal,scratched text and so on,it is shown that the algorithm can effectively reduce the time complexity and the probability of mis-matching.In this paper,the algorithm is judged by the combination of subjectivity and objectivity.The results show that all two improved algorithms have good repair efficiency,it verify the feasibility and rationality of the algorithm furtherly.