Abstract
We have previously (Wexler & Mamassian, VSS 2014) reported that there exist two independent biases related to the perception of 3D shape and motion from optic flow. These biases are both robust within observers, and highly variable across observers. Here we present a quantitative analysis of time series of these two variables, sampled once a day over three months in 97 observers. The temporal power spectra and autocorrelation structure of these time series show that they do not consist of independent samples from fixed distributions ("white noise"), but rather depend on internal variables that undergo cumulative changes over time. This observation is amplified by an analysis in the Box-Jenkins ARIMA framework, which shows that in most observers the time series are well modeled as random walks (or "Brownian motion") with superimposed measurement noise. Thus, we have evidence that the perception of at least two families of visual stimuli is governed by internal variables. Further, we show that (1) these variables can change both in response to external stimuli and perturbations, as well as during periods of complete darkness, showing that at least some of their dynamics is internal in origin; (2) the biases are strong enough to withstand a fair amount of counter-evidence from other depth cues; (3) the variables may not be single-valued but are actually vector fields over the visual field, with non-trivial spatial and temporal structure; (4) there are at least two more independent internal variables for 2D motion, corresponding to steps of 1D gratings and the motion quartet. The dynamic behavior of these variables opens a window on the internal dynamics of the visual system.
Meeting abstract presented at VSS 2015