Frames as abstractions for efficient multimedia object retrieval

Abstract

Multimedia Information Systems (MMISs) integrate various media types - text, audio, video, graphics, and animation - for machine-processable storage, retrieval, and presentation. These systems are characterized by complex logical interrelationships among multimedia objects, which, if efficiently represented as physical links, can significantly enhance retrieval performance. Traditional database models, including relational and object-oriented approaches, lack the expressive power needed to model such relationships effectively. This dissertation introduces a novel abstraction mechanism using frames, a concept from artificial intelligence, to model multimedia objects. Frames provide a rich interconnection structure and serve as flexible data models for representing stereotyped situations. A series of frame-based algorithms for multimedia object retrieval are proposed and evaluated through simulation. Results show substantial improvements in response time compared to traditional indexing methods. Further enhancements include a priority-based retrieval algorithm, which assigns request priorities based on time spent in the system, reducing response time variability. Another proposed algorithm prefetches related frames into memory, anticipating user needs and significantly increasing system throughput. Simulation studies demonstrate that the frame-based approach improves both performance and predictability. For example, average response time is halved, and system load capacity increases by an order of magnitude. The dissertation concludes with a summary of findings and outlines directions for future research in frame-based multimedia modeling.