Correspondence Noise Limits to Global Motion Perception

We measured human observers' thresholds for detecting coherent motion in random-dot kinematograms and compared them with theoretical thresholds predicted by the hypothesis that unavoidable correspondence noise in the kinematograms themselves is the limiting factor. In experiments where the stimulus had two fields we varied the dot density, the ratio of the number of dots in the two fields, and the area of stimulus. In further experiments we varied the number of fields and the method of generation of the coherent dots. The observed thresholds varied in the way predicted by the hypothesis over ranges that were in some cases wide, but even within these ranges the absolute levels of performance were much worse than that calculated for the ideal observer.

Objects moving in the real world generate motion signals in their images covering broad ranges of direction and velocity, and when the spread of motion signals within each presentation of our stimulus was increased, the discrepancy between ideal and actual performance was much reduced. We conclude that correspondence noise present in the stimulus is an important limiting factor in the coherent motion paradigm. Since this noise is largely external and inextricably bound with the signal, to attain high signal/noise ratios the input signals have to be pooled over broad ranges that match the range of motion signals in the natural stimuli. This provides an insight into the design of the motion detecting system, and perhaps into the design of other parts of the cortex.