The Fast Extraction Method Of Prospects And Prospects For Video Image

Matting is an operation in digital image processing. The target of Matting is extracting the area in an image where people have interest in, in order to change the background of this image or do some other operations. So what matting to do is deciding each pixel belongs to background, or foreground, or the mixture of background and foreground. We call the attribute of a pixel as alpha.However, the solving of alpha is an ill-problem, we cannot find a certain solution, or find a criterion to judge a solution is good or not. So it brings a big challenge to alpha matting. From Matting proposed till now, a lot of matting algorithms have been developed, some of them have reach the satisfactory matting effect. But there is still a problem in all algorithm, which is the calculation is too large. This problem limits the using scenarios of matting.In order to improve the speed of matting, we proposed Quick Matting and Neural Network Matting. Quick Matting focuses on image matting. In this method, we combine sampling calculation and neighboring calculation in one step, and we use the relationship between pixels alpha value. Therefore, compared with traditional methods, our method is25times quicker with the same matting effect. Neural Network Matting focuses on video matting. In the method, matting is treated as a learning problem as alpha is also regarded as membership in learning. Neural Network matting has good result and quicker and traditional matting methods. We use Neural Network Matting in real-time video matting which also achieve a good result. We also propose a process of real-time video matting, in which user only need to shake their body instead of drawing input.