Research On The Algorithm Of Video Segmentation

Image segmentation is a process of dividing digital image into several semantic areas, is one of the fundamental problems in computer vision direction. In recent years, with the development of multimedia, Internet technology, video as a medium for the transmission of information plays an increasingly important role in social life. Video segmentation began to rise and played a very important role in video coding, anomaly detection, target monitor and so on.Over the past few decades, in order to solve video segmentation problem, a large number of scholars have proposed a number of algorithms from different angels, which can be divided into semi-automatic algorithms and automatic algorithms in accordance with the need of artificial instructions, this paper focus on automatic segmentation algorithms. In a recent article, some scholars pointed out that, the current video segmentation algorithms mainly focuses on the texture, color, contour and other low-level attributes, but less space correlation analysis and multi-body motion characteristics between objects.With this research status, based on the traditional segmentation algorithm, a new unsupervised automatic video segmentation is proposed, which more focus on the correlation between objects.This algorithm represents the moving foreground with superpixel, reducing subsequent time, space complexity. Introducing the concept of link, match, in order to build the link weight model, turn the segmentation problem into a right calculation problem. Calculate the link weight, with static features from current frame associated with the relevance feature between frames, In order to optimize the search of relevance match between superpixels from different frames, the algorithm introduces superpixel color feature constraint and movement constraint.The experiment contains two aspects. The algorithm ensures higher recall rate and stable precision rate in the simple scenario, and complete single person segmentation from the crowd in the complex scenes. Large numbers of experiments show that the proposed algorithm can realize video image segmentation, and effectively solve the problem of over-segmentation.