A Model of Visual Attention for Natural Image Retrieval

Abstract

In this paper, saliency textons model is proposed to encode color, orientation and saliency cue and spatial information as image features for CBIR, where the image representation is so called saliency textons histogram. Experimental results indicate that the performances of saliency textons histogram outperform Gabor filter and multi-text on histogram. The saliency textons histogram can combine color feature, edge feature and spatial layout together. Furthermore, saliency textons model can simulate visual attention mechanism.