Image Restoration Method Based On Prior Information Learning And Priority Optimization

Digital image restoration refers to the repair of small-scale or large-scale damaged image,that is,filling the information defect area in the image,restoring the information of the defect area,and making the repair result look like it has never been missing or no repair trace.Image restoration technology has a wide range of applications in digital heritage protection,medical imaging,virtual reality,game production,and post-production of film and television.Research on digital image restoration technology has not only important theoretical significance,but also important practical application value.This thesis focuses on digital image restoration technology.An improved sparse representation image restoration method based on similar block group is proposed for small-scale damaged images.Firstly,neighborhood priori information is introduced to construct image block groups with similarity characteristics as training sample sets;secondly,similar block groups are trained to obtain adaptive multi-learning dictionary;finally,the image to be repaired is reconstructed based on adaptive multi-learning dictionary to achieve the purpose of image restoration.For large-scale damaged images,a fast image restoration method based on priority optimization with structure tensor is proposed.This method optimizes the priority function which determines the filling order in Criminisi method,makes the filling order more reasonable,and improves the efficiency of the algorithm significantly.The experimental results show that the improved sparse coding image restoration method based on similar block group makes full use of the prior information of the image,and obtains dictionary with strong expressive ability,which effectively improves the quality of image reconstruction.The work of this thesis is mainly embodied in the following three aspects:1.Emphasis is put on image restoration based on sparse representation.Considering making full use of the neighborhood information of the image to be repaired,this thesis proposes building similar block groups based on variance constraints,building similar block groups based on structural constraints and building similar block groups based on component constraints,further training multi-learning dictionary with adaptive features,and applying it to image reconstruction to be repaired.The experimental results show that the adaptive multi-learning dictionary has strong expressive ability and can effectively improve the quality of image restoration.2.The repairing quality of large-scale damaged images is closely related to the filling order.For example-based Criminisi method,the repairing results are mismatched and the efficiency of the algorithm is low.In this thesis,structural tensor optimization is introduced into the priority function in the Criminisi method of optimize the filling order of the repaired area.The experimental results show that this method not only effectively improves the quality of repair,but also significantly improves the efficiency of the algorithm.3.A digital image restoration system is designed and implemented by using MATLAB GUI.The system mainly includes two parts: system interface and algorithm function module.Function modules include image restoration algorithms for large area,small area and occlusion.