Research And Application Of Adaptive Image Completion Algorithm

As one of the hotspots in the field of image processing and computer vision,image completion technology has been widely used in cultural relics protection,film/television production,old photo restoration,and target removal,etc.Its core idea is to repair the damaged region using the known texture and structure information in image according to a certain repair sequence and matching criteria,and then to ensure the visual integrity of image.Firstly,the thesis introduces the background,research significance and current situation of image completion.And then,some popular algorithms are listed.Besides,we summarize the process,classification and characteristics of image completion,as well as the main problems and challenges in this field.Secondly,the advantages and disadvantages of various image completion algorithms are analyzed and summarized.Current exemplar-based algorithms lack of adaptability due to the manually determined patch size.Moreover,it is not suitable to obtain the optimal effect by using the same patch size in different structure and texture regions.In addition,the minimum absolute difference square sum(SSD)criterion,which is commonly used to search for the best matching patch in most exemplar-based algorithms,is simple and has limited accuracy,so it is not suitable to deal with the images with rich information with it.To address this problem,we propose two algorithms.1)To measure the structural difference,gradient information is added to the traditional structure similarity and is combined with the brightness,contrast ratio,and patch structure.On this base,the optimal size of patch in different structure or texture regions can be determined.2)To decrease mismatches and improve the accuracy of searching the optimal patch,the adaptive weight Census transform matching criterion is adopted.An improved adaptive weight Census transform matching criterion is introduced and combined with the classical SSD criterion.Experimental results show that the completion effect can be effectively improved for natural images with complex texture or gradually-changed color.Finally,the design principle of the image completion system is detailed.In the system,we integrate various image completion algorithms,as well as the scratch detection algorithms.Using human-computer interaction,any appropriate image completion algorithm can be selected,thus the completion process becomes more practical and convenient.