Image Inpainting Algorithm For Structure Damage Based On Global Statistical Information

Digital image inpainting technique is one of the research focuses in the computer vision.This technology has extensive application in the aspects of mural restoration,image compression,object removal.This thesis focuses on exemplar-base image inpainting.In order to maintain the structure coherence and reduce the time complexity of the algorithm,this thesis designs two improved algorithms.An exemplar-base image inpainting algorithm based on global statistics is designed.This algorithm matches similar patches and obtains their offsets.These offsets are used as labels.Candidate exemplar patches are only selected by labels.This eliminates interference,maintain the structure coherence and reduce the time complexity of the algorithm.The priority function is constructed by the structure tensor feature and structure sparsity.The image structure is partially restored and the image restoration order is more stable.The experimental results show that the repair results maintain the structure coherence.This proves the validity and progressiveness of improved algorithm.A global image inpainting algorithm based on structure tensor and color features is designed.The structural tensor and color features are introduced into the data item to maintain the structure coherence.The structural tensor and color features are introduced into the smooth item to maintain the neighborhood consistence.The quality of image restoration is improved by data item and smooth item.The experimental results show that the improved algorithm significantly improves the average PSNR at least 3dB and obtains higher image restoration quality.