Automatic Relevance Feedback for Distributed Content-Based Image Retrieval

In this paper, we present the machine-controlled relevance feedback technique for the distributed contentbased image retrieval (CBIR) system. A nonlinear model based on the Gaussian-shaped radial basis function (RBF) is applied in the feedback process, and a bias weighting is introduced to the query content as the partial supervised function to improve the retrieval precision. This paper introduces a decentralized Peer-to-Peer CBIR algorithm which reinforces offline feature calculation technique to generate a distributed feature descriptor database (DFDD), to offload feature computation to the P2P network while improving the retrieval precisions. In addition, this paper compares the retrieval performance over centralized, clustered, and decentralized peer-topeer network topologies. Combination of the ARF technique and the distributed CBIR system eliminates the human intervention, hence automates distributed CBIR in a hierarchical manner.