Analysis of vector space model and spatiotemporal segmentation for video indexing and retrieval

Region-based video indexing systems have opened up new possibilities for the description of visual content. However, these systems are affected by spatial variations on the regions obtained from image segmentation algorithms and by the complexity of region matching techniques. In this paper, we propose to enhance these systems with the use of spatiotemporal regions. The indexing framework studied for that purpose is based on the Vector Space Model (VSM), which enables efficient and compact shot representation with count vectors. We analyse the properties of the VSM and show that shot description can be improved by considering spatio temporal representations. For evaluation, we further compare the performance of the system using the spatiotemporal and the keyframe approach. Experimental results show that the spatiotemporal approach is advantageous in terms of retrieval performance and robustness of the description.