Noise reverses the oblique effect: A horizontal effect in orientation estimation and subjective uncertainty

The oblique effect refers to the classic finding that sensitivity to orientation information is worse for oblique (i.e., diagonal) compared to cardinal (i.e., horizontal and vertical) orientations. This textbook phenomenon is often taken to demonstrate our brain’s adaptive coding principles in response to environmental regularities. The oblique effect, however, is typically studied with stimuli that contain very narrow-band orientation content (such as lines or grating stimuli) which contrasts with most natural scene statistics, which are more broadband in nature (i.e., they contain a variety of orientation content). Surprisingly, several papers have found a reversal of the oblique effect using detection tasks and stimuli filtered to include more broadband orientation content, a finding dubbed the horizontal effect (since detection is usually worse for horizontal, followed by vertical, then oblique stimuli). Here we show that simply adding pixel noise to a grating stimulus is sufficient to induce the horizontal effect. In a large sample, using an orientation estimation task combined with subjective uncertainty judgments, we found a clear increase in both errors and uncertainty ratings for horizontal stimuli and a relative reduction of error and uncertainty for oblique stimuli, in direct contrast to the standard oblique effect. We also found that increasing the contrast of the noise in the stimulus seems to induce a larger orientation-related modulation of errors. Pixel noise is a common manipulation used to degrade stimuli yet our results suggest that doing so also fundamentally alters the sensitivity to orientation content in the image. Moreover, the dependency of the horizontal effect on noise contrast could suggest an explanation of the phenomenon in terms of divisive normalization.