The Tilt Illusion and the Orientation Bandwidth of Primary Level Filters

Abstract

Models for pattern discrimination can be applied successfully to simple orientation discrimination tasks. However, the mechanisms underlying long-range spatial effects, such as the tilt illusion, are not well understood. The magnitude of the tilt illusion and its sensitivity to the polarity of the inducers were measured by means of tests with varying orientation bandwidths. The orientation tuning bandwidth of the primary-level filters, whose peak frequency is 1.7 cycles deg⁻¹, was estimated in a 2AFC orientation-discrimination task. The luminance distribution of the stimulus was jointly normal. First, the orientation bandwidth of a single elongated test stimulus was increased by reducing its length-to-width ratio from 7.5:1 to 1.5:1. The orientation discrimination performance rapidly decreased after the ratio became less than 3.5:1, which is in good agreement with the estimates of the shape of the filters (Wilson, 1986 Vision Research 26 453 – 469). These optimally shaped stimuli were used as inducers in the tilt illusion experiment.

Six 20° tilted inducers were added around the test stimulus at the radial distance of 1.25 deg. When the length-to-width ratio of the test stimulus was varied as in the first experiment, the magnitude of the tilt illusion increased after the ratio was less than 3.5:1. However, the precision of the orientation discrimination when compared to the first experiment did not change. Changing the polarity of the inducers did not affect the illusion. The results suggest that the tilt effects of the inducers are not mediated by the primary-level filters. Rather, strengthening of the tilt illusion as the orientation bandwidth of the test stimulus increases suggests that long-range interactions between second-order orientation mechanisms might underlie the described phenomena.