Abstract
When dynamic visual noise is viewed with an interocular delay, it appears in depth. Near depth planes appear to stream horizontally from the undelayed to the delayed eye, while far depth planes stream in the opposite direction. This illusion is often cited as psychophysical evidence for the joint encoding of disparity and motion. In this view, the percept is supported by neurons which are sensitive to the direction of horizontal motion as well as to binocular disparity. This explanation was strengthened by the observation of neurons with the appropriate properties in cat area 17/18. However, recent studies have found that they are rather rare in monkey V1. Motivated by this, we examined whether the percept can also be explained even if disparity and motion are initially encoded separately rather than jointly. It has been argued that a similar explanation, proposed by Tyler, cannot explain several experimental phenomena. However, these failures have never been demonstrated in a fully implemented model. We constructed a population of model neurons, which were either pure motion sensors (sensitive to direction of motion but not to disparity) or pure disparity sensors (sensitive to disparity but not to direction of motion), and simulated its response to dynamic visual noise with an interocular delay. We found that the activity of motion sensors tuned to horizontal motion in the direction of the delayed eye was correlated with the activity of near disparity sensors, while the activity of sensors tuned to the opposite direction of motion was correlated with the activity of far disparity sensors. We suggest that this correlation is sufficient in itself to explain the depth percept in all the tested configurations. We conclude that the depth illusion reflects spatial disparities present in the stimulus, and so is consistent with any reasonable neuronal mechanism of stereo depth perception. It therefore does not provide specific evidence for joint motion/disparity encoding.