Abstract
Perceiving the 3D world has been modeled as a process of probabilistic inference that produces noisy depth estimates from depth cues. In this account, the Just-Noticeable Difference (JND) is used as a proxy measure of cue noise and is measured as the minimum depth difference that can be reliably discriminated between a fixed standard stimulus and an experimentally varied comparison stimulus. This interpretation is based on two assumptions: (1) The slope of the perceptual function relating physical to perceived depth is unitary and (2) perceptual uncertainty yields noisy depth estimates. Here we show that these two assumptions are wrong. Based on a new theory of 3D processing we hypothesize that (1) the slope of the perceptual function is not unitary but depends on the strength of the depth cue; (2) depth perception is deterministic and does not yield noisy depth estimates; (3) the noise affecting discrimination is task related and independent of depth processing. According to this new theory, it is the slope of the perceptual function that determines the JND: if the slope is small, it requires a large change of the physical depth to bring about a change in perceived depth that can be discriminated. The JND is therefore entirely dependent on the experimentally varied comparison stimulus, since the standard stimulus is fixed and yields a noiseless depth estimate. In two experiments, we manipulated the strength of a Structure-from-Motion (SFM) stimulus by varying the 3D rotational speed and studied discrimination between it and a disparity defined stimulus. Remarkably, we found that the JND only depended on the comparison stimulus: faster SFM stimuli and disparity stimuli cause smaller JNDs than the slower rotating SFM stimuli. Conversely, the JND was completely independent from the nature of the standard stimulus, providing strong evidence of a deterministic theory of 3D cue processing.