For humans and other primates there are few perceptual tasks more important than visual search. Visual search is performed almost continuously during waking hours, and is essential for locating and interacting with objects, places and other organisms. Loss of either central or peripheral vision can seriously degrade search performance. The overall aim of this proposal is to understand the visual mechanisms underlying search for the normal real-world case where multiple fixations occur. We propose a multifaceted and unique attack on this difficult problem. Our first specific aim is to develop a quantitative characterization of the responses of primary visual cortex neurons (in monkey) to the transient variations in local mean luminance and contrast that are typical of those occurring when the eyes are engaged in multiple-fixation search of the natural environment. Our second specific aim is to develop a quantitative characterization of the responses of visual cortex neurons to transient presentations of target stimuli embedded in backgrounds of l/f noise (which is representative of the amplitude spectra of natural images) and in backgrounds of white noise (which has been used extensively in psychophysical studies of visual detection and visual search). There is almost no previous work on these aims, yet they are critical for developing and testing models of visual search performance. Our third specific aim is to measure multiple-fixation search performance in humans under carefully controlled conditions for targets embedded in l/f and white noise, and for targets in natural scenes. To do this we will use gaze-contingent software developed in our lab, that allows precise real-time control of the content of a visual display relative to the observer's current gaze direction, as measured with a precision eye tracker. In these experiments, we will measure and quantify the role of central and peripheral vision on search time, accuracy, and eye movement patterns. The fourth specific aim is to develop quantitative models that will be tested against the results of our gaze contingent display experiments. Our starting point will be an ideal Bayesian observer of multiple-fixation search for targets embedded in broadband noise backgrounds. This ideal searcher shows how to optimally integrate information across the visual field and across fixations, and how fixation locations should be selected; it also provides the appropriate benchmark against which to evaluate real search performance. Testable models of visual search will be formulated by degrading the ideal searcher in various ways; including, for example, incorporating constraints based upon the neurophysiological experiments in the first two aims, and by proposing constraints on visual memory and information integration within and across fixations.
Showing the most recent 10 out of 33 publications