Digital Image And Video Restoration Technology Research

Digital image and video completion aims at filling in missing pixels in anunknown region of an image or video in a visually plausible way, and it has a wideapplication area. Since a video consists of a series of image frames, we could treat avideo as a video cube. Under this assumption, we could extend image completiontechnologies, to the video completion fields. In this paper, we discuss image and videocompletion in a utility framework. Image and video completion is one of the hottestareas in computer vision, and a great deal of improvement has been achieved.Firstly, we summarize the existing algorithms on image and videocompletion, and introduce various algorithms based on PDEs or Texture Synthesis.Secondly, we analysis the features of images and videos, and introduce algorithms foridentifying various features. Thirdly, we present a new exemplar-based framework forimage and video completion. We develop the Markov Random Fields(MRF) forimages and videos, based on which we regard video completion as a discrete globaloptimization on a multi-labeling problems. Then the.optimization is solved by a novelalgorithm, called Smart BP, which alleviates the problems of slow convergence andintolerable storage size when using the standard BP. Furthermore, we use avision-related pixel weight for various feature descriptors, which make the completionresults more appealing to human vision. Finally, we give a conclusion about our paperand present some future work.All these methods have great potential in many applications, especially in moviespecial effects, image and video completion, virtual reality, image and videocompression and so on.