Modeling the Interactions of Bottom-Up and Top-Down Guidance in Visual Attention

We propose a framework of visual attention grounded on earlier attentional and perceptual models to serve as a basis for the development of computational vision systems. The framework is build on generally established knowledge about the neural basis of human attention, models developed by Briand [1][2], Itti and Koch [3][4], and Wolfe [5][6] and our own results that call for adaptations of the existing models [7][8]. In this paper we concentrate on the interaction of bottom-up and top-down processes to understand the mechanisms underlying exogenous and endogenous attention. Two series of studies are reported to support the proposed adaptation of the earlier models. First, we claim that the visual feature binding is a common process for exogenous and endogenous attention [7]. Secondly, we demonstrate the ability to preset the bottom-up feature maps by demonstrating the phenomenon of Inhibition of Return (IOR) with endogenous cueing, suggesting that IOR affects processing before focusing attention [8].