Video Semantic Annotated Automatically Based On Bayesian Inference

With the development of computer network and multimedia technology, video data grow with the unprecedented speeds. However video information management technology only can provide non-linear,non-intelligent operation for customer, which cannot satisfy the need of access and management of vast and rich video information, massive information can not be reused. It becomes a problem to be solved how to realize automatic or semiautomatic video information management and provide convenient information management system. Under such background, Content-Based Video Analysis and Retrieval (CBVAR) technology is proposed to analyze,organize and manage the vast video data, CBVAR is a effective solution for the above conflict and becomes hot point in the research field of multimedia technology now.The low-level visual features of video are static feature including color, texture, and shape, etc. But the description and understanding of video for people are mainly on semantic level, complexity, fuzziness and subjectivity of video semantic content lead "semantic gap" between low-level feature and high-level feature, moreover this problem haven't been solved very well. Therefore how to get video semantic annotation automatically becomes a vital topic.In this paper, the existing video semantic annotation technology is deeply analyzed and a method is proposed to annotate non-fixed length annotation for videos. First, Bayesian network is established to express the co-occurring relationship of semantic concepts. Then, Na?ve Bayesian classifier is used to obtain the semantic candidate set of video. Finally, Bayesian Inference is used to get the final annotation set from the semantic candidate set. The experimental result indicates the proposed method of annotating is efficient.