Abstract
Transparent motion presents a major challenge for the motion-processing system, as multiple global directions must be detected within the same region of space. Some computational models have resolved this by extracting the perceived directions from multiple activity peaks across the global-motion population. However, recent electrophysiological evidence suggests that population activity becomes unimodal when the angular separation between component directions is below 90° (Treue et al., 2000). Were this the case, more complex operations would be required to detect transparency with acute angular separations. Alternately, the averaged population response may not reflect the specific activity used in transparent-motion detection, particularly if the visual system relies on a subset of detectors with narrow direction-tuning bandwidths. We thus sought to determine the population activity produced by near-threshold transparent-motion stimuli using psychophysical techniques. Following adaptation to transparent motion, population activity was assessed through the elevation of unidirectional detection thresholds for several test directions. Adaptation to suprathreshold and threshold-level angular separations produced a bimodal pattern of elevation in detection thresholds, with higher elevation for test stimuli moving in the component directions than the mean direction. These peaks were lost at subthreshold angular separations, with equivalent threshold elevation for detection of the mean and component directions. To quantify the height of these population activity peaks, transparent-motion adaptation was also compared with unidirectional adaptation at various coherence levels. With suprathreshold angular separations, threshold elevation for the component directions was equivalent to unidirectional adaptation with 50% coherence. Threshold elevation for subthreshold angular separations approached that produced by 100% coherent unidirectional motion. Thus, subthreshold angular separations produce population activity resembling unidirectional motion in both shape and magnitude. Together, this suggests that bimodality is an essential requirement for transparent-motion detection, with peaks determined by global-motion signal-to-noise processes.
Supported by an Australian Research Council grant (#S6505064).