Conceptual Indexing for Video Retrieval

There is a growing demand for technology that will allow computers to better manage and manipulate video and other media sources. We are building a system to help create and search an archive of stock video clips in order to lower the cost of new video and multimedia productions. The goal is to provide on-line access to stored video clips with flexible, effective, and efficient retrieval. Our approach to this retrieval problem is to construct a rich conceptual indexing system, a simple retrieval algorithm, and an easyto-use browsing interface. Because the current state of the art in video analysis does not support automated extraction of many interesting conceptual features from video, we focus our efforts on elaborating a vocabulary of what those features should be and on building a system in which it is practical to hand-code indexes. We argue that the conceptual indexing approaches of existing case-based reasoning systems in AI can be adapted to meet the representational requirements of intelligent multimedia retrieval systems. We present six categories of conceptual indexes that are applicable to stock video clips, argue for our approach to the development of effective end-user indexing and retrieval systems, and describe our current implementations.