Hugh R. Wilson - US grants

DESCRIPTION (The applicant's abstract): Previous research has firmly established that the retinal is initially processed in parallel by a number of spatial mechanisms that are selectively tuned for orientation and spatial frequency. Although these parallel channels have frequently been treated as independent, it is now clear that this independence reflects the relative simplicity of stimuli used in previous experiments rather than true mechanism independence. Accordingly, the overall goal of this proposal is to determine how interactions among spatial mechanisms contribute to the processing of complex visual information. This goal will be pursued in three major areas: visual analysis of two-dimensional motion, psychophysics of contrast gain control, and mechanism interactions in disparity pooling and stereopsis. In all three areas experimentation will be complemented by network modeling studies designed to explain the data and provide a guide to the design of further experiments. The studies of two-dimensional motion will utilize stimuli comprised of sums of cosine gratings at different orientations and moving in different directions. Previous research funded by this grant has demonstrated that the direction of motion of many of these patterns cannot be predicted accurately by the intersection of constraints construction. Two key goals will therefore be to extend this analysis to motion in the peripheral visual field and to determine the conditions under which the sum of two moving gratings can capture the motion of a third grating of different spatial frequency, orientation, and velocity. A powerful new masking technique develop under this grant has provided evidence for the existence of a division operation that adjusts the response level of visual mechanisms so as to keep them within an optimal processing range. The characteristics of this contrast gain control will be explored as a function of the orientation and spatial frequency of the masking grating. In addition, its role in binocular vision will be ascertained by splitting the mask between the two eyes. Research on stereopsis will determine the spatial frequency and orientation ranges over which disparity pooling occurs. Stereograms will be sums of gratings of two different spatial frequencies and orientations, with each grating presented at a different disparity. In addition to the use of these stimuli to measure disparity pooling, they will also be used in binocular masking studies that are linked to the contrast gain control experiments. As the experiments in these three areas are interrelated, common principles of spatial mechanism interaction should emerge. Thus, this research will greatly expand our understanding of the manner in which mechanism interactions aid in processing complex visual information.

Recent research has shown that inherently nonlinear mechanisms are required to explain many aspects of texture discrimination and motion perception. These mechanisms all incorporate oriented filtering followed by rectification and then a second oriented filtering stage and are termed "non-Fourier" or "second order" processes. The proposed research will test the hypothesis that higher level form vision likewise utilizes non-Fourier shape mechanisms. In this proposal a non-Fourier mechanism capable of processing quasi-circular and ellipsoidal contours is developed, and networks of these units are shown to be capable of analyzing the shapes of curved objects. These non- Fourier form mechanisms also agree with characteristics of concentric units found in cortical area V4 in primates (Gallant et al, 1993) and with V4 functional anatomy (Schoups et al, 1995). Pilot data involving discrimination of circles and Glass (1969) patterns support the existence of these mechanisms in human vision by showing that information is summed linearly along curved contours. Major goals of this proposal are therefore: (1) measure the characteristics of non-Fourier form mechanisms psychophysically; (2) use these mechanisms to predict Glass (1969) pattern perception; (3) gather psychophysical data and use the same mechanisms to predict discrimination for a very general mathematical class of shapes; (4) extend these results to stereopsis; and (5) determine whether other nonlinear models, including that of Kovacs and Julesz (1993, 1994), might explain the data. The study of non-Fourier form mechanisms offers the promise of a new leap forward in our understanding of higher form vision. The principles learned should lead to insights into the perception of human faces and prosopagnosia, into strabismic amblyopia, and may soon permit us to optimize image proceeding systems for analyzing medical images to detect tumors, etc.