| The overwhelming influx of digital information to which the world is subjected has resulted in vast, and ever-growing repositories of multimedia data. Furthermore, the practicality of using centralized techniques for tracking and indexing such huge volumes of data is becoming resource intensive. Moreover, centralized schemes suffer from the fact that they demonstrate complete system loss upon experiencing component failures. To address these problems, this thesis discussed the application of social acquaintance networking theory for creating an open architecture for the distributed indexing of multimedia data, and specifically, images.; The proposed indexing model employs the sociological theory of ' Small Worlds', and mirrors the directed nature of connections seen in the World Wide Web. In applying this model to multimedia, images transcend their static nature as database entries, and become active participants in a self-organizing, distributed, peer-to-peer network. The self-organization of these small world nodes (images) is established according to the distances between their MPEG-7 description data, and results in all images maintaining 'links' to a fixed number of highly similar peers. Furthermore, in order to promote distributivity, each participating node is required to store feature description(s) of both itself and of its peers. For scalability, a network growth algorithm is proposed to facilitate the introduction of new small world nodes into the system. This growth algorithm exploits the fact that existing network nodes locally store peer description data, thus allowing new nodes to obtain referrals to images which successively exhibit descriptions that are increasingly similar to their own.; A similar referral strategy is also employed for performing image retrievals by enabling a Small World retrieval agent to converge upon data that matches user-specified search criteria. This retrieval agent, however, serves a much greater purpose as it is used to illustrate the proposed indexing technique's resiliency to component failures. Experiments demonstrate that as the number of failures introduced into the Small World index increases, the viability of performing image retrievals is maintained, but with a correspondingly degraded performance. This graceful degradation is useful in mission-critical situations where a loss of service is unacceptable, and slightly less-than-optimal operation can be tolerated until component failures can be corrected. |
