Abstract
Content-based video indexing and retrieval has its foundations in the analyses of the prime video temporal structures. Consequently, technologies for video segmentation and key-frame extraction have become crucial for the development of advanced digital video systems. Conventional algorithms for video partitioning and key-frame extraction are mainly implemented autonomously. By focusing the analysis on compressed video features, this paper introduces a real-time algorithm for scene change detection and key-frame extraction that generates frame difference metrics by analysing the statistics of the macro-block features extracted from an MPEG compressed stream. The key-frame extraction method is implemented using difference metrics in curve simplification by means of a discrete contour evolution algorithm. This approach resulted in a fast and robust algorithm. Results of computer simulations are reported.