Feature-based saliency for a model of bottom-up visual attention

Abstract

Visual attention poses mechanisms for the selection of salient stimuli from visual scenes. The aim of this paper is to model the first phase of a three-phase mechanism of bottom-up visual attention. This phase considers feature-based saliency for attentional deployment. Feature-based saliency is computed by combining two saliences: the first saliency considers simple visual attributes: intensity, color and orientation; the second one takes into account object's shape. The stimulus with the greatest feature-based saliency is attended by making an eye movement to bring it to the center of gaze or by improving its contrast sensitivity. The proposed model is intended to serve as a basis for complex visual experiments performed by virtual creatures, which are endowed with a cognitive architecture.