The Effect of Gabor Element Aspect Ratio on Contour Detection
John S. Nafziger1, Shih-Cheng Yen2, and Leif H. Finkel1,2
1Institute of Neurological Sciences
3Department of Bioengineering
3320 Smith Walk, 301 Hayden Hall
University of Pennsylvania
Philadelphia, PA 19104, U. S. A.
Abstract
Contour detection thresholds have previously been determined by varying inter-element distance or the difference in orientation between elements (Field et al., 1993; Kovács et al., 1996). Kovács et al., 1996 have shown that the contour detection threshold depends upon the relative spacing of contour and background elements rather than upon absolute inter-element distance. We investigated the local mechanisms of contour detection as a combined function of orientation, element spacing, background density and aspect ratio. Subjects performed a 2AFC task to detect contours embedded in stimuli composed of 144, 256, or 400 randomly oriented Gabor patches extended over a viewing area of 17° x 17°. Individual Gabor patches were 0.4° x 0.4° (5.0 cpd). Contour detection thresholds were determined as a function of the ratio of contour element spacing over background element spacing (ø-1). Performance curves were generated for contours defined by a constant inter-element orientation change. The threshold element spacing at each orientation was plotted in polar coordinates producing a graphical representation of an element's local connection field. Small inter-element orientation changes (0° - 20°) allow a greater inter-element integration distance than large orientation changes (>40°). The local connection field was scale invariant; its shape was not modified by changes in background element density. Increasing the Gabor element aspect ratio significantly increased the inter-element threshold spacing, supporting that the recruiting of additional cells and/or sharpening of the orientation response extends the range of contour integration.
(Supported by ONR, NSF and The Whitaker Foundation)