A “back-pocket” model of V1 orientation processing predicts curvature blindness

Poster Presentation: Tuesday, May 20, 2025, 2:45 – 6:45 pm, Banyan Breezeway
Session: Spatial Vision: Models

Steven Dakin1 (), Selassie Tagoh2; 1University of Auckland, 2University of Auckland

Takahashi’s popular “curvature blindness” illusion (2017, i-Perception 8(6): 1–12) reveals that modulation of luminance along a sinusoidal-shaped line can dramatically alter its perceived shape, (notably) causing smooth contours to appear to have sharp corners. We asked how well a simple model of V1 orientation processing might predict this striking illusion. We presented a series of test stimuli: sinusoidal lines whose luminance was modulated along the line according to a square wave with a similar period to the sinusoid but with a variety of phase offsets (compared to the line shape). Six observers then adjusted the shape of a larger, white sinusoidal reference lines to match the appearance of a given test. Observers did this by adjusting the amplitude and “sharpness” (0: pure sinusoid, 1: pure triangle wave) of the reference. The sharpness could be adjusted independently for the portions of the reference to the left and right of its peak (allowing us to capture perceived asymmetry). As well as replicating the original finding – that luminance transitions at the peaks/troughs of the curve (90 deg phase offset) make it appear much sharper – we report that these transitions increase the perceived amplitude of the sinusoid (by at least 25%). Stimuli with 45 and 135 deg phase-offsets induce asymmetry in sharpness around the peaks and troughs of the test patterns. A simple “back-pocket” model of V1 orientation processing– using oriented Gabor filters operating in a winner-take-all manner at a given spatial location - predicts perceived changes in sharpness, asymmetry and amplitude measured in the matching experiment. This powerful shape illusion emerges naturally from V1 processing without the need for dedicated explanations involving e.g. segmentation.

Acknowledgements: Supported by the Leitl Charitable Trust