Abstract

Although an edge is defined by an abrupt change of luminance, the luminance profile of its retinal image affects its perceived location. We measured the trade-off between luminance and perceived location of an edge by interposing a thin, uniform strip of variable luminance between the black (0.7 nit) and white (70 nits) sides of a sharp edge and having observers make vernier adjustments of an uncomplicated sharp edge, so that the two edges appeared to be aligned. We found that the perceived location of the contour depended on the luminance of the strip, even when the strip was not detectable. The relation between perceived location and luminance was linear for strips 1 ft. (30.4 cm) wide or less but nonlinear for wider strips. Transfer of information was maximum at 1.4 ft. (42.7 cm), the widest strip that could not be detected at any luminance. A model involving three stages of spatial filtering and an interposed nonlinearity described the data from strips that could not be seen, with no free parameters. With the optimum strip width, 1.4 ft. (42.7 cm), the model had errors of prediction and residual variance that were half of those produced by models based on the assumption of linearity.

© 1990 Optical Society of America