Research On Video Matting Algorithm

The matting is a technology of extracting interested part from the whole image. In film production, photo processing and virtual reality field, it has been widely used. How to accurately extract the interested part of image or video has been emphasis and difficulty in matting field. The natural image matting technology has no limitation of the images and it also can obtain good matting effect, so it is used very conveniently and attracts more and more attention. The early natural image matting need complex manual calibration to get Trimap, and run slowly. Some of recent natural matting methods reduce the human interaction, but they couldn’t process the image with complicate colors. Therefore, in this paper, the aim of study is to improve the efficiency and effect of matting by improving the existing matting algorithm and try to introduce the image matting into video matting.The aim of video matting is to extract the interested part from every frame, it need to matting for every frame. If we use natural matting method for calibrating every frame of video manual, we require a lot of processing time and it is hard to ensure space-time consistency of video matting. Aim at the above problem, in this paper, we mainly discuss the method of using interframe correlation to improve the efficiency and expect to reduce the time of video matting and ensure the space-time consistency of video matting.In this paper, the main work is as follows:1. Research on robust matting algorithm which is based on sampling and transmission, this paper proposes a method of reducing matting time by using the adaptive window to reduce the number of unknown pixels which are need to be computed the mask values. First we calculate the mask estimation value of a pixel in unknown area and its corresponding optimal foreground and background samples pair and the samples pair’s reliability, by the local smoothness of image, we can learn that several pixels around the pixel also corresponds to the optimal sample pair, so it can greatly reduce the calculated number of unknown pixels and improve the efficiency of matting; This paper also analyses the weight problems in the process of getting the unknown pixel’s mask valve, the improved algorithm fully considers the gray information’s influence on the mask estimation valves and increases a gray weight on the basis of distance ratio and Euclidean distance weight. The experimental results show that, improved matting algorithm ensures the higher matting’s quality and also shorten the matting time.2. Research on the algorithm of getting keyframes of a video, this paper proposes a method of getting video keyframes by combining sports ground energy and dichotomy, this method uses the dichotomy to segment video, by comparing the energy field of the middle frame and keyframe to determine the middle frame is keyframe or not. Experiments show that this method can greatly reduce the calculation time of keyframes and the keyframes can better represent the whole video content.3. This paper also analyzes and studies the transfer of keyframes’ Trimap. After obtaining the keyframes and their Trimap, we pass the keyframes’ Trimap to the follow-up frames by using the correlation between video frames. In this paper, we adopt a Trimap transfer strategy based on the gray information, by comparing the gray value differences of adjacent frames’pixels in the same position to determine which Trimap areas the pixels belonging to, such successful pass Trimap to subsequent frames. Experiments show that this method can achieve better balance of effectiveness and efficiency.