Image clustering on peer-to-peer network

A two-stage image clustering method is presented in the paper in order to decrease transmission data on the peer-to-peer (P2P) network and improve image clustering efficiency. We define the concepts of peer, class, group and overlay, and describe the architecture of P2P network. The image clustering method includes two stages: intra-clustering and inter-clustering. In the intra-clustering stage, images on a peer are clustered into classes. We propose multiway spectral clustering algorithm with kernel 2-directional 2-dimensional principle component analysis and group images in one peer. In the inter-clustering one, the feature space is divided into units into which the feature vectors may be mapped, thus images can be represents by a simpler manner according to the partition units. We propose class similarity and group similarity algorithms, and give the peer clustering strategy. Finally, the experiment results show that the method reduces transmission content on the network while increase the performance.