Research Of Certain Target Removal And Image Inpainting

Accompany with the fast development of modern computer technology, storage medium and net technologies are becoming more and more improved. A mass of digital images are saved in all sorts of storage mediums and transimited on the net. Digital image processing technology is under fast development due to its vast application in many aspects. Image target removal, as a branch technology of the former, is also known as image inpainting. It aims to remove one or several specified targets in images and reconstruct the missing area using the still remaining image information. The process seeks to make the dealed image credible and natural without visual conflicts. Image target removal is respected to have broad underlying application fields.In this paper image inpainting methods are sorted into two classes: pixel-level image inpainting and patch-level image inpainting. Different ideas and inpainting mechanisms of the two classes are compared while several classical inpainting algorithms are deeply analyzed. For pixel-level image inpainting, BSCB Model, Total Variation Model and Fast Marching Inpainting Method are taken as examples, while for patch-level inpainting, Texture Synthesis Technologies, local optimization patch-exemplar method and global optimization patch-exemplar method are given as examples.Several inpainting methods including BSCB Model, Fast Marching Method and local optimization patch-exemplar method are discreted into numerical models and programmed. Merits and disadvantages of these algorithms are compared through a good many of experiments in which different targets with defferent backgrounds are removed. Thoeretical explaination is given to interpret the experimental results. Different application situations proved suitable to the algorithms are also discussed.A new searching strategy is proposed to improve the patch-searching-step in local optimization patch-exemplar method. The new strategy avoids to searching match patches in the whole image, instead, it locates image region which has characteristic similar to the current patch and searchs the best matching patch only in this region, thus ,the searching becomes more effective. In addition, a new application framework is also proposed to carry out target removal based on the patch-exemplar method. Experiments show that in this framework computer is well conducted by the user during the inpainting procedure to get satisfying results.