A Local Model for Transparent Motions Based on Spatio-Temporal Filters

Psychophysical studies suggest that humans perceive moving multiple sine patterns as transparent when the underlying component gratings are of sufficiently different speed, spatial frequency, orientation, and contrast. We propose a gradient inhibition model that explains these phenomena using only local intensity or contrast information. Computational results show a contrast dependent transition for splitting/ coherence of sine plaid images. This model suggests that some transparency information can be locally computed and subsequently fed into global vision modules.