Knowledge-Supported Segmentation and SemanticContents Extraction from MPEG Videos for Highlight-Based Annotation, Indexing and Retrieval

摘要

Automatic recognition of highlights from videos is a fundamental and challenging problem for content-based indexing and retrieval applications. In this paper, we propose techniques to solve this problem by using knowledge supported extraction of semantic contents, and compressed-domain processing is employed for efficiency. Firstly, video shots are detected by using knowledge-supported rules. Then, human objects are detected via statistical skin detection. Meanwhile, camera motion like zoom in is identified. Finally, highlights of zooming in human objects are extracted and used for annotation, indexing and retrieval of the whole videos. Results from large data of test videos have demonstrated the accuracy and robustness of the proposed techniques.