Background Replacement For Hand-Hold Video

Video background replacement is an important operation in computer vision, usually used in film and television special effects. So it always has an important research significance. However, the process is always completed in a studio, with specialized equipment and person, making is difficult to use for common users. This paper is in order to break through the restriction of the traditional method, providing a convenient way to make special effects for home users.In order to complete convincing background replacement, it needs to overcome several non-trivial challenges. First, the foreground performance should be segmented from its original video, along with an alpha matte for each frame. Second, the camera motion of the new compositing video and the foreground performance must be matched, eliminating the incorrect visual effects caused by inconsistent camera motion. Finally, the lighting of the foreground object and the new background must be made compatible and consistent.As to the problem of natural video segmentation, we can extract the foreground subject quickly with the interactive tools based on local classifiers. To the problem of inconsistent camera motion, this paper presents a method called dynamic space-time warping, ordering and warping the selected target background frames in a manner that optimizes a well-designed objective function, effectively removes the artifacts caused by mismatched camera motion.This paper first introduces some related work to video background replacement briefly. Then describes a simple method based on image stitching and gives some discussion to its limitations. Finally explains dynamic space-time warping in detail. After evaluating on a various of examples, we get a good performance on speed and effectiveness.