Edge co-occurrence in natural images predicts contour grouping performance.

The human brain manages to correctly interpret almost every visual image it receives from the environment. Underlying this ability are contour grouping mechanisms that appropriately link local edge elements into global contours. Although a general view of how the brain achieves effective contour grouping has emerged, there have been a number of different specific proposals and few successes at quantitatively predicting performance. These previous proposals have been developed largely by intuition and computational trial and error. A more principled approach is to begin with an examination of the statistical properties of contours that exist in natural images, because it is these statistics that drove the evolution of the grouping mechanisms. Here we report measurements of both absolute and Bayesian edge co-occurrence statistics in natural images, as well as human performance for detecting natural-shaped contours in complex backgrounds. We find that contour detection performance is quantitatively predicted by a local grouping rule derived directly from the co-occurrence statistics, in combination with a very simple integration rule (a transitivity rule) that links the locally grouped contour elements into longer contours.

Direction biasing by brief apparent motion stimuli.

The perceived direction of a motion step (probe stimulus) can be influenced by an earlier motion step or a brief motion sweep containing a series of steps (biasing stimulus). Depending upon experimental conditions, the biasing of the direction of the probe step (a phase shift of 180 degrees +/-Phi) by a biasing stimulus which precedes it by approximately 250 ms can either increase (positive filter biasing) or decrease (negative filter biasing) the tendency to see the probe move in the biasing direction as computed with a motion filter with a biphasic temporal impulse response. In a series of experiments it was found that biasing motions traversing 90 degrees of phase angle in fewer than six steps in less than 100 ms produced positive filter biasing. Also, biasing of the probe direction could be dissociated from the consciously reported direction of the biasing stimulus, and it did not occur when the probe preceded rather than followed the biasing stimulus. A biasing sweep containing more than six steps traversing 90 degrees or a sweep traversing 270 degrees produced negative filter biasing. Perceptual fusion of the steps of the sweep was not a necessary condition for obtaining negative filter biasing. In general, the negative filter biasing effects were found to be the most pervasive for the conditions investigated, and they are suggestive of a direction-specific, adaptation-like (gain-control) process in first-order motion filters. The exception to the negative biasing rule was found only with biasing stimuli which were short in duration or distance spanned.

Computational analyses in cognitive neuroscience: in defense of biological implausibility.

Because cognitive neuroscience researchers attempt to understand the human mind by bridging behavior and brain, they expect computational analyses to be biologically plausible. In this paper, biologically implausible computational analyses are shown to have critical and essential roles in the various stages and domains of cognitive neuroscience research. Specifically, biologically implausible computational analyses can contribute to (1) understanding and characterizing the problem that is being studied, (2) examining the availability of information and its representation, and (3) evaluating and understanding the neuronal solution. In the context of the distinct types of contributions made by certain computational analyses, the biological plausibility of those analyses is altogether irrelevant. These biologically implausible models are nevertheless relevant and important for biologically driven research.

Viewer-centered temporal biasing of 3-D rotation percepts.

The perceived direction of rotation of a 3-D cloud of dots can be biased by a prior rotation (Jiang, Pantle, and Mark, 1998 Perception & Psychophysics 60 275-286). In a series of experiments, it is shown that the temporal rotation bias is reversed by a 180 degrees change of head orientation between two rotation sequences; i.e. the perceived direction of rotation reverses for the second of two sequences when head orientation is changed. The bias is, therefore, viewer-centered. Perceptual reversals are not obtained when the orientation of the head is changed and returned to its original position between rotation sequences. It was also found that the viewer-centered bias combined additively with viewer-independent near-far luminance information. Finally, the bias was manifest when 3-D depth was re-established, but not maintained, between rotation sequences. A model, in descriptive and flowchart forms, is used to explain the integration of world-centered information and a viewer-centered temporal bias on the presence/absence of perceptual reversals of the rotating virtual sphere. In the model, the temporal bias is the result of the coupling of depth values to persisting 2-D retinal motion signals.