Abstract
Normal observers can detect extended contours defined only by long-range correlations in orientation, but contour detectability decreases as the level of background noise increases. In a psychophysical task, spatial noise could act by decreasing the signal from the contour elements (masking) or by increasing uncertainty as to which elements are relevant. To discriminate between these two views, we have used a direct measure of signal strength, the Visual Evoked Potential (VEP), to obtain thresholds for Gabor-defined contours. Circular contours (n= 7−11) comprised of 12 Gabor patches were presented at random locations in the presence of noisy backgrounds comprised of identical but randomly oriented Gabors. Contour elements were rotated on and off the implicit contour at 3.0 Hz and noise elements were rotated at 3.6 Hz. By tagging the contour and noise elements with different temporal frequencies, we were able to record contour responses in the presence of varying degrees of noise without contamination of the contour-element responses by responses to the noise. Noise tolerance thresholds were measured for different degrees of orientation offset at a fixed spacing of the contour elements or at a fixed orientation offset for a range of contour element spacings. The addition of high-density noise eliminated the contour-specific response even though all the contour elements were still present and were unchanged in form. Contours whose elements had 0–10 deg orientation offsets were most resistant to noise. Contour-specific components were measurable at contour-element spacings up 9 wavelengths of the Gabor carrier (1.6 deg). The decrease in contour-specific responses with added noise is not consistent with uncertainty models. These reductions in signal strength are more easily explained by a masking effect. Thus explanations that rely exclusively on the observer being uncertain in the presence of noise are unlikely to be correct.