Content-based image retrieval with automated relevance feedback over distributed peer-to-peer network

In this paper, we present a novel content-based image retrieval (CBIR) system over a distributed peer-to-peer (P2P) network, with the goals to (1) reduce the computation requirement for content similarity matching and (2) provide a higher retrieval precision. The performance gain is achieved by utilizing the network infrastructure and user behaviors. By identifying the community neighbors from nearby peers, all the subsequent content queries are made within the community neighborhood, hence the feature matching process is reduced and a higher retrieval precision is achieved. Under the distributed P2P infrastructure, we investigate the CBIR system with a nonlinear relevance feedback based on Gaussian-shaped radial basis function (RBF) with partial supervision. To minimize the level of human interaction, we automate the relevance feedback process by polythetic clustering technique with rank-weighting and hierarchical decision-making using self-organizing tree map (SOTM). Simulation of the proposed system shows that a larger community neighborhood achieves higher retrieval precision.