Bayes-Based Relevance Feedback Method for CBIR

Abstract

The paper proposes a Bayes-based relevance feedback approach integrating visual features and semantics for content-based image retrieval systems. The data of the image database are divided into small clusters by semantic supervised clustering algorithm. The cluster here is called as index cluster. So the data of each index cluster are similar both in visual features and in semanitcs. During relevance feedback process, users sign the positive and negative examples regarded a cluster as unit rather than a single image, and each feedback cluster construct a Bayessian classifier on visual features; and the semantic classes of the feedback examples construct the Bayessian classifier on semantics. At last, we use Bayesian classifiers on visual features and semantics respectively to adjust retrieval similarity distance. Our experiments on an image database show that a few cycles of relevance feedback by the proposed approach can significantly improve the retrieval precision.