Research On Digital Image Inpainting Algorithms

Digital image inpainting is the technique whose purpose is to restore a partially damaged or missing image to resume its integrity and original visual effect.It is of great importance in a variety of applications,such as protection of cultural relic and paintings, restoration of damaged images caused by network transmission,removal of caption, scratches and target objects in images and videos etc.Researching status of image inpainting is summarized,and some representative inpainting algorithms and their advantages and disadvantages are analyzed in this paper. On this basis,two inpainting algorithms are proposed:(1) Image inpainting algorithm based on adaptive templates and updating of confidence term.This algorithm uses adaptive templates to search and match the restored area,that is to say the sizes of these templates alter to local character of the image adaptively.When updating the confidence term,in order to avoid "error match", this algorithm adopts negative exponential function to make the updated confidence value be inversely proportional to the "cumulative error".The experimental results show that this algorithm can effectively improve the inpainting results and reduce "garbage objects" caused by "cumulative error".(2) Image inpainting algorithm based on fractal theory.This algorithm introduces fractal theory into the field of image inpainting and restores damaged images according to the self-similarity principle.Firstly,the source region of the image is partitioned into a set of blocks which form a codebook.Secondly,the best-match block is chosen from the codebook by the way of fast fractal,so that the pixel values of the region to be filled do not rely on only the copies of the existing region,which can resolve the problem that the searching area is limited to the source region of the image.In addition,in order to intensify the constraint of searching condition,the proportion of confidence term is increased while computing the priority value,thus the filling process proceeds by the order of "onion-peel",while linear structures propagate into the target region simultaneously,so that "false edge" caused by "error match" can be avoided because of the filling order.The experimental results show that the proposed algorithm has very good performance in aspects of both perceptual quality and inpainting efficiency,and still has a good effect for the damaged images whose loss ratio is relatively large.