Bayesian modeling of perceived surface slant from actively-generated and passively-observed optic flow

Area of Science:

• Visual Perception
• Computational Neuroscience
• Psychophysics

Background:

• Perceived surface slant is crucial for depth perception and navigation.
• Existing models of active vision offer different explanations for slant perception, particularly regarding the use of head motion cues.

Purpose of the Study:

• To investigate how perceived surface slant is affected by optic flow generated during active (head movement) and passive (simulated) viewing.
• To compare the predictive accuracy of an "inverse optics" model (using head motion) versus a probabilistic model (ignoring head motion) for slant perception.

Main Methods:

• Measured perceived surface slant from optic flow for physical and virtual objects under different head movement speeds (normal, slower).
• Simulated optic flow on a computer for stationary observers to isolate the effects of visual input.
• Compared empirical results with predictions from Bayesian "inverse optics" and probabilistic models.

Main Results:

• Perceived surface slant was systematically distorted in both active and passive viewing conditions.
• Distortions increased with head translation speed, correlating with the optic flow's local velocity gradient.
• Empirical findings aligned with the predictions of the probabilistic model, not the "inverse optics" model.

Conclusions:

• Perceived surface slant in active vision may arise from probabilistic processes that do not ensure veridical perception.
• Extra-retinal signals (head motion) might primarily aid in refining retinal information processing rather than directly correcting slant estimates.