Image Completion Using Sample Synthesis

摘要

The issue of image completion is well developed in these years. Most of them reconstruct damaged area by referring to the under-repaired images themselves. However, they may fail if damaged portion collapse their structure which is unique and important. We propose to use an external reference image to repair the damaged image and the method provides the following three contributions: (1) an algorithm of fast contour matching is proposed to repair damaged images by referring to external sample images. Critical structural information can be rebuilt which is missing in original damaged area. (2) An algorithm of fragment color transform is proposed to resolve the problem of creating false transform to non-existent color if traditional histogram specification were used. (3) An algorithm of K map threshold weighted synthesis is proposed to resolve the problem of creating false textures caused by non-existence of similar block in original damaged area. Several experiments are executed and the results clearly indicate that defects mentioned above are able to fix more efficiently. Especially, the present method shows good performance for texture accommodation in the joint area. Therefore, it is ideal for the task of completing images with unique structure missing in the damaged area.