• Content-based video representation enabling scalable browsing and summarization through hierarchical summarization, clustering, and scene-transition cues across media streams and compressed video [14], [15], [6], [4], [20], [8].

• Temporal modeling for action recognition and tracking, leveraging time-sequential features, hidden Markov models, and dynamic motion analysis to achieve robust recognition in varied camera conditions [7], [19], [16], [11], [10].

• Efficient video indexing and editing through compressed-domain analysis: abrupt scene-change detection in MPEG streams and motion-vector based indexing enabling fast retrieval and navigation [3], [8].

• Discovery of underlying story units and narrative structure to enable non-linear browsing of long programs via content and temporal cues [2], [5], [18].