Perception at equiluminance: an adaptive model of motion metamers

When the perceived intensities of a moving color pattern are equalized the pattern will seem to slow down or stop moving. This situation creates motion metamers - two different motions can be perceived as equivalent. Motion metamers present a very difficult problem for classical models of machine motion perception. In this paper we present an adaptive filter model of motion interpretation which describes human motion perception performance at equiluminance. This model strongly suggests that the interaction between the pathways is defined by a specific structure and the learning parameter between the pathways are different. This result makes several predictions about the human motion processing and places some performance limits on machine vision models.