The object-based selective attention has been given increasingly importance in psychology domain in recent years

whereas most of the existing models of selective attention in computer vision field are either feature-based or space-based.In this paper

an object-based computational model of selective attention is proposed where "perceptual object" is postulated as the basic attention unit.The proposed attention model consists of following two steps:(1) how to select the first focus of attention in a given image;(2) how to shift the attention within the whole image.Under this model

the contrast

defined as the absolute gray difference between the "perceptual object" and its neighborhood

is used as the measure of the object's salience

by which the attention is determined and shifted.The advantages of this model lie in: Firstly

the model is entirely based on perceptual objects

its results can be easily applied to object detection

image segmentation

and scene analysis.Secondly

the model is of multi-scale

flexible enough to be easily adjusted according to the specific applications.Extensive experiments on real images show that the proposed attention model works reasonably.