Abstract
Human motion sensitivity is typically measured using artificial displays such as random dot kinematograms that provide close control over stimulus parameters but poorly approximate natural input. We developed a technique that allows measurements of sensitivity to motion components using movies of natural scenes. Using a 2-AFC match-to-sample paradigm, we investigated the tolerance of the visual system to speed noise for movie stimuli. Individual frames of grayscale 1-second movies of natural scenes were analyzed using derivative-of-Gaussian filters that defined local orientation, contrast and spatial frequency, and were then reconstructed using a subset of these filters. On each successive frame elements were tracked locally based on distance and similarity, creating a measure of cumulative displacement through time. Speed noise could then be added by altering the displacement of each element. A noiseless source movie was presented for 1 second, followed by a reference movie that contained a fixed amount of speed noise and a target movie containing an additional amount of speed noise under control of a staircase procedure. There were six levels of speed noise from 2.5% increase to 80% increase in logarithmic steps. Subjects were asked to identify which of the simultaneously presented movies were more similar to the source movie. A bootstrapping procedure identified the threshold for 75% correct identification. Subjects showed a monotonic increase in thresholds with the addition of pedestal noise that was significantly fit by an equivalent noise function (t = 13.44, p < .001). Previously we reported finding a dipper function in a similar pedestal paradigm using an orientation discrimination task with static images and suggested that this function arises from dependence upon a higher-level perceptual template that defines the location of expected features. The present results suggest a difference in the encoding of static orientation and speed information for moving and static natural scenes.