Separating Global Motion Components in Transparent Visual Stimuli – A Phenomenological Analysis

Abstract

When two distinct movements overlap in the same region of the visual field, human observers may perceive motion transparency. This perception requires the visual system to separate informative and non informative motion signals into transparent components. In this study, we explored the computational constraints in solving this signal separation task - particularly for the stimulus configuration where two grating components move in the same direction at different speeds. We developed a phenomenological model which demonstrates that separation can be achieved only for stimuli with a broadband Fourier spectrum. The model identifies the informative component signals from the non informative signals by considering edges. This approach is shown to be limited by an edge sensitive spatial filtering of the image sequence, the threshold tolerance for local signals considered and the number of iterative computational steps.