Constructing semantic network based on Bayesian Network

There is more and more video information on the Web. The recognition of semantic information from visual content is an important task in video retrieval. Semantic network which captures the semantic relationships among concepts can be used for video annotation. In this paper, we present an improved three-phase dependency analysis (ITPDA) algorithm constructing Bayesian network to automatically discover the relationship network among the concepts, and then we can use the constructed semantic network to annotate an unknown video shot. The advantage over the traditional three-phase dependency analysis (TTPDA) algorithm is that no requirement for the users to provide any node ordering. The system can automatically orient the edges of the network when users can not give a node ordering. The computation complexity is reduced from O(N⁴) to O(N²) (N is the number of nodes in the network) when orienting the edges. Experimental results show that ITPDA performs better than TTPDA algorithm in the application of automatic semantic video annotation.