Spatial Structure From Temporal Structure

In these displays there is no spatial structure defined by luminance, contrast, orientation, or direction of motion. It is only the temporal synchrony of changes in motion direction that sets off one region from another. In the top pair of animations point processes describing changes in direction of motion of the gabor elements in the "figure" are perfectly correlated - 1.00; in the lower pair of animations, these "figure" point processes are correlated 0.50. In all four of these animations, point processes for Gabor patches in the "background" region have an average correlation of 0.00. In the two left-hand animations, the "figure" is taller than it is wide ("vertical rectangle") while in the two right-hand animations the "figure" is wider than it is tall ("horizontal rectangle"). In our experiments we vary the aspect ratio of this correlated region to measure the accuracy of shape perception from temporal synchrony. NOTE: Video frame-rate influences the vividness of figure/ground segregation, and these animations are slower than those used in our experiments - if you have trouble seeing the rectangle, try moving the slider manually at different rates. Also, use the right-hand arrow (->) to step through the animation frame-by-frame, to convince yourself that the form is not visible in individual frames or, for that matter, in two successive frames.

 

"Vertical" 100% correlation

 

"Horizontal" 100% correlation

 

"Vertical" 50% correlation

 

"Horizontal" 50% correlation


In the animation shown below, point processes among Gabor elements defining the "figure" are perfectly correlated (1.00) among themselves, and point processes defining the "background" are perfectly correlated (1.00) among themselves. But, the two sets of point processes - those defining the figure and those defining the background - are uncorrelated (0.00). If anything, figure/ground segmentation is even more vivid in this condition.


Return to text