Abstract
In the absence of reliable and abundant depth cues, estimates of surface slant are often biased towards fronto-parallel. Here we investigate the effects of curvature distortions on perceived slant. In general, curvature distortions are predicted to decrease the precision of slant discrimination, and this uncertainty may, in turn, strengthen the fronto-parallel bias. Alternatively, curvature distortion might bias slant perception independently, or in the opposite direction, of the fronto-parallel bias. We rendered images of slanted, textured surfaces with and without radially symmetric distortions (pincushion and barrel) at low (≈1%) and high (≈5%) levels. Observers judged whether a test image (distorted or undistorted, with a variety of slants) was more slanted than a distortion-free surface with a 15° slant. We fit the psychometric data with a cumulative normal function and estimated bias and discrimination thresholds for each observer. Our results showed that 1% distortion had no measurable impact on slant discrimination. At 5%, both types of distortion significantly increased slant discrimination thresholds. However, only the pincushion distortion produced a systematic underestimation of perceived slant. Slant underestimation in the presence of pincushion distortion is consistent with the hypothesized effect of disparity smoothing operations. Under this hypothesis, slant should also be underestimated in the barrel distortion condition but it is not. To test the possibility that this type of curvature distortion introduces additional perceptual biases, in ongoing experiments we are measuring perceived slant magnitude in the presence and absence of curvature distortion. These suprathreshold estimates will provide a baseline for the fronto-parallel bias in isolation; additional biases in the distortion conditions could then be modelled as distortion-based effects.
Acknowledgement: NSERC grant CRDPJ 523632-18, CFREF Vision: Science to Applications program