An Integrated Approach for Content-Based Video Object Segmentation and Retrieval Part II: Content-Based Searching of Video Objects

In the previous part of this paper, we presented an active object segmentation system which can be used to extract semantic video objects with accurate boundaries. New video representations based on these semantic objects provide great potential for content-based video searching. In this part, we present a new content-based video object searching system. The system uses a novel query model which effectively combines a comprehensive array of visual features at both the object and region levels, as well as matching of various spatial-temporal structures. Experiments have shown encouraging results and great promise for future advanced video search tools.