Research On Technology Of Image Completion Based On Texture Synthesis

With the popularization of digital recorders, a large number of digital images are produced, but digital images are easy to be damaged. How to filling in damaged images is a widespread problem. This thesis mainly research on digital image completion, which is a work of filling in missing or damaged portions of the image according to known image information. Its objective is to make the completion non-detectable for an observer who does not know the original image. Image completion is widely used in many applications, and is a new hotspot in image processing.In the research of image texture completion, the thesis proposes an adaptive patch matching algorithm based on structural similarity for the block effect problem of patch-based texture synthesis. Firstly, the algorithm chooses patch matching method and the size of patch according to texture features. Then the algorithm considers pixel error, texture structural error and boundary error to improve the structural similarity between matching patch and template patch, thus reduces block effect. Experimental results demonstrate that the proposed algorithm can improve texture synthesis quality effectively for irregular texture image.In the research of image structure completion, to resolve the structure confusion brought by greed mode image completion, the thesis proposes a novel image completion approach based on structure identification and automatic structure completion. The approach first gets salient structure information on the basis of image segmentation, and analyses structure features according to curvature and symmetry. Then the approach uses lines and Bezier curves to complete non-closed missing structure and uses shape matching to complete closed structure. Finally completes the image by patch-based texture synthesis according to completed structure curves. Experimental results demonstrate that the proposed approach can well solve complex structure natural image completion.On the basis of previous study, an objective image completion quality measure (ICQM) is proposed. Firstly, curvature similarity between completed structure and known structure is used to assess structural completion quality. Then texture distribution, texture repetitiveness and texture block effect are used to assess texture completion quality. Finally, ICQM is proposed according to assessment result of structure and texture. Experimental results demonstrate that ICQM is an effective assessment index which has high correlation with human subjective assessment result.