IBM Research TREC 2002 Video Retrieval System.

Abstract

In this paper, we describe the IBM Research system for analysis, indexing, and retrieval of video, which was applied to the TREC-2002 video retrieval benchmark. The system explores methods for fully-automatic content analysis, shot boundary detection, multi-modal feature extraction, statistical modeling for semantic concept detection, and speech recognition and indexing. The system supports querying based on automatically extracted features, models, and speech information. Additional interactive methods for querying include multiple-example and relevance feedback searching, cluster, concept, and storyboard browsing, and iterative fusion based on user-selected aggregation and combination functions. 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. In this paper, we describe the approaches for each of the tasks and discuss some of the results. 1