Perceptual Science Series
Human performance predicted by optimal processing of natural image movies
Dr. Johannes Burge
Monday, November 17, 2014, 12:00pm - 07:00pm
University of Pennsylvania, Department of Psychology
Accurate perception of motion is essential in sighted animals for locomotion and for detection of predators and prey. Further, accurate motion perception depends critically on accurate estimation of motion speed in the images formed by the eyes. Here, we first analyze the statistical structure natural image movies to determine the optimal receptive fields for units encoding local motion speed in a given direction. Next, from the receptive field responses to natural stimuli, we determine the neural computations that are optimal for combining and decoding their responses into estimates of speed. The computations show how units with selective, invariant speed tuning might be constructed by the nervous system. Finally, we compare human and optimal speed discrimination performance on the same set of naturalistic stimuli. Human speed discrimination performance closely parallels optimal performance. Indeed, with a single efficiency parameter, the detailed shapes of a large set of human psychometric functions are accurately predicted by the optimal computations. We conclude i) that many properties of speed selective neurons and of human speed discrimination performance are predicted by the optimal computations, and ii) that natural stimulus variation affects optimal and human observers almost identically.