Small world distributed access of multimedia data: an indexing system that mimics social acquaintance networks

Abstract

This paper proposes a technique employing the concept of small-world theory to achieve an acquaintance network made up of various types of media objects. Mirroring the way in which humans keep track of descriptions of their friends and acquaintances, every media object within the Small World Indexing Model (SWIM) actively participates in storing descriptions of the objects that are most similar to itself. This results in an extremely high level of decentralization, where each object participates as an equal member in a peer-to-peer network and no central index is required. Retrieval within this ubiquitously networked environment is performed using an agent-based technology exploiting similarity between query criteria and node-specific descriptions stored locally by each media project. This framework is extremely general in that it can easily be applied to any multimedia data type and also modified to employ any low-level or semantic descriptor