| The increased availability and usage of multimedia information stored in digital repositories have created a critical need for efficient multimedia processing, retrieval, and transmission algorithms. In this thesis, we introduce a novel content-based video retrieval and tracking system (VORTEX), with functionalities that facilitate the automated processing of multimedia information. A crucial step in multimedia processing is that of reliable video segmentation into visually coherent video shots (scene change detection). We present a new one-pass, real-time approach to video scene change detection based on the statistical sequential analysis theory and operating on compressed multimedia bitstream. Furthermore, we develop an unified framework for effective detection of both abrupt and gradual scene changes. Object tracking and retrieval from compressed multimedia bitstreams enables the hierarchical processing of the video contents and the formulation of queries in terms of semantic characterizations of visual activity. In this thesis, we introduce a novel approach that permits multiple object tracking from compressed multimedia databases. Finally, we present a new algorithm for the efficient representation of visual information contained in video sequences. The approach is based on the Principal Component Analysis and takes advantage of the locality of the data in video shots, and the optimal energy compaction properties of the transform. |
