Hunt the Cluster: A Scalable, Interactive Time Bayesian Image Browser for P2P Networks

In Peer-to-Peer (P2P) networks, computers with equal rights form a logical (overlay) network in order to provide a common service that lies beyond the capacity of each single participant. Among other applications, fielded P2P networks have shown their viability for the distribution and exchange of large amounts of data. Current research on retrieval in P2P systems focuses largely on keyword-based retrieval and other weakly in- teractive query paradigms. Within this paper, we present a Bayesian image browser that helps the user in finding images in dis- tributed collections. Bayesian image browsers operate by presenting sequences of thumbnail-based collections to the user, at each step collecting user feedback that is used to update a Bayesian model. In contrast to query by example (QBE) no initial query image is needed in order to start the query process. Our approach is scalable in that the state of the Bayesian model is maintained locally in the browsing peer and only a small number of thumbnails is requested from the network at each step. Each query step is thus done in a short time frame. In this paper we present the method, as well as first experiments done using a JXTA-based implementation.