Contours drive the tilt aftereffect in naturalistic images

Orientation is a fundamental property of visual scenes, yet real-world images often contain multiple, sometimes conflicting, orientation signals. Sharp object boundaries, shading gradients, and fine surface textures can each suggest different orientations in the same region, forcing the visual system to decide which signal to prioritize. The tilt aftereffect (TAE)-a robust illusion in which prolonged exposure to a tilted stimulus makes a near-vertical target appear tilted in the opposite direction-provides a sensitive probe of how orientation is encoded and adapted. Across six experiments, we dissociate orientation defined by extended contours from orientation energy captured by standard filter-based models and test which of these signals drives the tilt aftereffect when they conflict. Experiment 1 replicates the classic tilt aftereffect (TAE) using Gabor patches. Experiment 2 introduces natural-scene patches in which contour-derived and filter-derived orientations were put into direct conflict and showed that adaptation followed contour orientation even when filter energy indicated the opposite tilt. Experiment 3 links this bias in adaptation to explicit orientation judgments on the same images. Experiment 4A and B replicates the effect using orientation estimates from human-traced line drawings to increase contour fidelity, and Experiment 5 tests whether blocked exposure strengthens natural-scene TAEs toward the magnitude observed with simple stimuli. Together, these findings indicate that contour structure, rather than undifferentiated orientation energy, plays a privileged role in how the human visual system encodes and adapts to orientation in complex scenes.