Noise Limits in the Computation for Motion Detection

Extracting motion information from visual signals is a crucial function for the survival of most animals. A widely accepted model of how the visual system detects wide-field motion is built on computing correlations in the input spatiotemporal light pattern sampled at separate locations in space and time. I used psychophysical experiments to measure the limits of the spatial span and time delay over which the human visual system correlates these input signals. These results set the parameters in a correlation-based ideal observer model. Ideal observer performance is compared to human performance on a psychophysical task based on the estimation of motion direction. The comparison between these results give insight into how robust this computation is to input noise either in the form of contrast fluctuations or photon shot noise.