Abstract
Aim. Several models of curvature detection assume that curvature is detected by mechanisms that compare local luminance-defined orientations along the contour. Here, we tested the alternative notion that the inputs to curvature detection mechanisms are local contour positions. Method. Contours consisted of strings of oriented Gabor elements with various curvature frequencies. The curvature of the contour could be defined in terms of Gabor position, Gabor orientation, or both. In the position-only condition Gabor orientation was constant, and in the orientation-only condition the Gabor positions formed a straight line. In a 2-IFC procedure subjects were to determine which of two contours was curved. Threshold curvature amplitudes were determined for each of the three conditions. Results. Curvature amplitude thresholds were similar for the position-only and position-and-orientation conditions. However, thresholds were elevated for the orientation-only condition, especially at low curvature frequencies. Conclusion. Our results suggest that the inputs to contour curvature mechanisms are the positions of local contrast-energy along the contour, with little or no input from luminance-defined orientations, particularly at low curvature frequencies.
Supported by NSERC (Canada) grant OGP 0121713 given to F.K.