Content-Based Image Retrieval Based On Visual Attention And The Conditional Probability

摘要

A novel content-based image retrieval framework was presented in this paper. This framework is used to encode primary visual feature and saliency information as natural image features by simulating visual attention mechanism and using the conditional probability. In this framework, the color volume is used as a novel feature to detect saliency areas. Besides, a novel generalized visual feature representation method, namely the conditional probability histogram, is proposed to describe natural image features. It can integrate primary visual features and saliency information into one whole unit. Experimental results indicate that the proposed algorithm outperform our prior works, namely multi-texton histogram and color difference histogram.