Combining cues while avoiding perceptual conflicts.

Abstract

A common assumption in cue combination models is that small discrepancies between cues are due to the limited resolution of the individual cues. Whenever this assumption holds, information from the separate cues can best be combined to give a single, more accurate estimate of the property of interest. We examined whether information about the discrepancy itself is lost when this is done. In our experiments, subjects were required to combine cues to match certain properties while avoiding perceptual conflicts. In part 1, they combined expansion and change in disparity to estimate motion in depth; and in part 2, they combined perspective and binocular disparities to estimate slant. We compared the pattern in the way that subjects set the two cues with the patterns predicted by models of cue combination with and without a loss of information about the discrepancy. From this comparison we conclude that little information about the discrepancies between cues is lost when the cues are combined.