Video retrieval using VQ-based global motion features

In this paper, we propose a Vector Quantisation (VQ)-based video retrieval algorithm using global motion features. The contribution includes two points: first, we design a VQ-based algorithm for eliminating the singular points in the motion vector space; second, we utilise the global motion vector index histogram of all frames in the query video clip to match those of all video clips in the database. To retrieval histograms instead of original motion vectors can reduce the time computation complexity, and thus assure a real-time video retrieval process. The experimental results show that the proposed algorithm can effectively extract the statistical characteristics of the global motion features. In addition, the computation complexity is low while the retrieval accuracy is high.