A re-ranking approach to video retrieval on the peer-to-peer network

This paper presents a re-ranking method for retrieval of video clips in a distributed database system. The method employs a peer-to-peer architecture to implement a network of distributed computing and storage of video files. The retrieval can be started by a query node forwarding query packets to its peer nodes, and followed by the pseudo-relevance feedback within all peers. The retrieval accuracy can be improved by the forward-backward propagation of signal between the query node and the peer nodes. This process is done automatically and has no requirement for the transmission of video files over the network during the pseudo-relevance feedback learning process. This reduces the required bandwidth utilization. Based on the experimental result, the proposed re-ranking method appears to be very effective for retrieval of video clips stored in the peer-to-peer database and converges in a few iterations.