A Saliency Map Model for Color Images using Statistical Information and Local Competitive Relations of Extracted Features
Kyung Joo Cheoi, Yillbyung Lee · 2002
Biological systems appear to employ a serial strategy by which an attentional spotlight rapidly selects circumscribed regions in the scene, rather than attempting to fully interpret visual scenes in a parallel manner for further analysis. In this paper, we propose a biologically motivated saliency map model and applied our system to locating candidate regions of interest on various images for further detailed analysis. In proposed model, several basic features are extracted directly from visual stimuli, and these features are integrated based on their statistical information and local competitive relations. Through integration process, unnecessary features for detecting the salient object are spontaneously decreased while useful features are enhanced. The performance of the model is evaluated over various color images of synthetic and complex real images.