Abstract
Visual attention is known to be involved in task control, yet not much is known about the role of attentional processing in different tasks. Attention is often investigated in visual search, where the task is to localize an object in a scene, while another basic task is to recognize an object, where the attentional processing is largely unexplored. However, we propose visual attention being involved in multiple tasks differently and investigate how visual attention might operate in several complex tasks. The motivation comes from the complexity of real-world tasks. Typical is a continuous interplay between different basic tasks, like first searching an object, identifying it, using it, etc. Based on a recent neuro-computational model of visual attention (Beuth, Doctoral thesis, 2019; Beuth & Hamker, 2015, Vision Research), replicating a large set of neurophysiological data, we selectively disabled different mechanisms that altogether define attention in the general case, and measured performance drops. Analysis was performed on an image data base (COIL-100) and a real-world application (wafer inspection). For visual search, we found that the top-down amplification of specific objects and features is important, as well as the spatial suppression between different places. While for recognition, we diametrically found a significance for the spatial amplification, and for the feature-based suppression. Hence, we observed opposing influences and found out that visual attention operates diametrically in the localization and recognition task in the visual system. In more complex tasks (Object Substitution Masking), composed of visual search and recognition, we found an interplay takes place between both mechanisms, according to the phases of the involved processes. These results illustrate how visual attention may operate in two major task classes, and the model predicts how complex tasks can be solved based on these 'building' blocks to realize the complex task nature of our real-world.