Community-based multimedia information retrieval in peer-to-peer networks

Abstract

With the rapid development of information technology, multimedia applications have been widely used in people's daily lives. The need of developing effective and efficient multimedia information retrieval technologies has been identified in recent years. Due to the large amount of computational power needed for the searching and processing of multimedia data, as well as the fact that multimedia data are often naturally distributed over a network, distributed multimedia information retrieval has attracted researchers’ attentions. The distributed multimedia information retrieval environment consists of autonomous computer systems that form a peer-to-peer (P2P) network. Each computer has a multimedia database installed and hosts some multimedia data in which the owner of the database is interested. The data stored in the database can be queried by the owner as well as other peers in the network. In order to reduce the number of peers accessed for processing a query, the routes of query forwarding should be carefully selected so that only relevant peers are visited. Existing distributed multimedia information retrieval technologies depend solely on peer content summaries for the clustering of relevant peers for route selection. This may lead to problems in heterogeneous environments. The lack of semantics inherent to these approaches may cause strange things to happen (e.g., funny results). Due to technical limitations, the content summaries provided by the source peers might be biased and cannot well reflect the true content in their databases. Furthermore, in credit rewarding content sharing systems for autonomous networks, peers have incentive to provide biased content summaries in order to gain improper priority to retrieve certain type of content. Another potential problem is that in some settings there are non-cooperative peers who are not willing to provide more information other than a simple query-response interface. The purpose of the current study is to develop a novel multimedia information retrieval framework for distributed networks, which is suitable for both cooperative and non-cooperative environments and resistant to biased content summaries. The relations of peers are established and evolved according to their historical interactions, which are computed with a peer reputation model. Peers sharing the same topic tend to group themselves together to form communities. We show that networks constructed in this way have the following characteristics: (1) peers tend to have direct links that point to others who have high retrieval qualities on the same topic, and (2) those that have the same topic tend to have shorter distance in the graph. The resulting structure of the network is exploited to facilitate distributed information retrieval. The characteristic of the network structure is used to infer the content of non-cooperative peers, as well as refining possibly biased content summaries provided by peers. A query forwarding algorithm is introduced for the routing of queries. Simulation experiments show good performance of the proposed method. The method is tested with a multi-agent system on a large image dataset for distributed content-based image retrieval (CBIR). Its performance is compared with existing approaches. The results of the simulation show that the proposed method achieves the shortest average distance after some time of the network evolving and the retrieval performance is getting better as time goes. Scalability experiments are also conducted to show the feasibility of the proposed method in large scale networks.

Date of Award	17 Feb 2010
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Qing LI (Supervisor) & Wenyin LIU (Co-supervisor)