Second-Order Processing of Visual Information

Research Overview

Non-Fourier (or second-order) motion stimuli, popularized by Chubb and Sperling (JOSA, 1988), are signals whose perceived motion is not predicted by typical energy or gradient-based models. These stimuli include the motion of contrast envelopes (e.g., shadows), occlusion boundaries, the motion of illusory contours, and some illusory motions caused by aliasing. In an attempt to provide a unifying formal account of these stimuli and their relation to models for non-Fourier processing, Keith Langley (University College London) and I have examined a framework that is derived, in part, from the definition of group velocity in wave mechanics. We showed that many non-Fourier stimuli, when viewed as multiplicative combinations of elementary signals, have simple descriptions in the Fourier domain.

I have also been involved in several psychophysical studies that are directly related to these modeling efforts. Langley, Paul Hibbard (University of Surrey), and I reported evidence supporting the hypothesis that contrast envelopes are processed by the visual system after orientation- and scale-specific filtering (in visual cortex). This showed that their perception is not an artifactual consequence of an early nonlinearity (e.g., in photo-transduction), which had been proposed to explain these percepts.

In binocular vision, Langley and I also found evidence for a non-Fourier process, much like that in motion analysis. In examining this issue in greater depth, Langley, Hibbard and I showed that non-Fourier stimuli sometimes produce a percept of transparency, distinct in its properties from transparent percepts that arise from a superposition of two signals. This supported the hypothesis of a non-Fourier channel in stereo depth perception.

One of our first hypotheses was that non-Fourier stimuli are related to distinct physical properties of natural scenes (e.g., multiplicative transparency and occlusions), and therefore non-Fourier processing channels may not subserve all the same visual tasks (such as egomotion) as conventional first-order models. However, Rick Gurnsey, Cindy Potechin (Concordia University) and I found that non-Fourier motion can be used to induce a percept of self-motion (vection). In this study we found a dissociation between motion-aftereffects (non-existent with non-Fourier stimuli) and vection as the relative amounts of Fourier and nonFourier motion energy were varied in the stimuli.

Related Publications

Return to David Fleet's home page.