Perceptual Grouping and Gelatinous Ellipses

Yair Weiss and Edward H. Adelson

email: yweiss at cs dot huji dot ac dot il


The ease by which we compute motion of objects in a scene is deceptive. Here are some demonstrations of the ambiguity of motion and our visual system's use of perceptual grouping in order to resolve this ambiguity.

Note: If you cannot see the embedded videos, you can download them using the links in the caption. They are Quicktime format and will play in any compatible player.

A narrow ellipse oscillating rigidly about its center appears rigid (quicktime 191k).
However, a fat ellipse undergoing the same motion appears nonrigid (quicktime 197k).
The apparent nonrigidity of a fat ellipse is not really a "visual illusion". A rotating ellipse or a nonrigid pulsating ellipse can cause the exact same stimulation on our retinas. In this sequence (quicktime 370k) the ellipse contour is always doing the same thing, only the markers' motion changes.
The ellipse's motion can be influenced by features not physically connected to the ellipse. In this sequence (quicktime 420k) the ellipse is always doing the same thing, only the dots' motion changes.
The ellipse's motion is not influenced by spurious features. In this sequence (quicktime 407k) our visual system parses the scene into two rigidly moving objects, rather than one nonrigdly deforming one.

So what does this mean?

Our visual system integrates multiple measurements in order to analyze the motions in a scene. It also makes a decision which measurements to integrate and which to segment. We have developed a computational model of motion integration that is consistent with human observers' perception of these stimuli. The model is consistent with a wide range of previously published phenomena and is described in this paper.

Perceptual Science Group Home Page