|
Abstract : |
In this paper, we describe the IBM Research system for analysis, indexing, and retrieval of video, which was applied to the TREC2002 video retrieval benchmark. The system explores novel methods for fully-automatic content analysis, shot boundary detection, multi-modal feature extraction, statistical modeling JOt semantic concept detection, and speech recognition and indexing. The system support, ' querying based on automatically extracted features, models, and *7eech inJbrmation. Additional interactive methods ' for querying include multiple-example and relevance feedback searching, cluster, concept, and storyboard browsing, and iterative fision based on user-selected aggregation and combinationfunction. s'. The system was applied to all four of the tasks of the video retrieval benchmark including shot boundary detection, concept detection, concept exchange, and search. We describe the approaches for each of the tasks and discuss some of the results. diverse methods for video analysis, indexing, and retrieval, which included automatic descriptor extraction, statistical modeling, and multi-modal fusion. We conducted experiments that individually explored audio-visual and speech modalities as well as their coinbination in manual and interactive querying. In the paper, we describe the video indexing and retrieval system and discuss the re-suits on the video retrieval benchmark., |
