The same problem arises for edges as for many of the other gestalt fragments and gestalts formed in the visual system. Since in certain settings the visual system assigns a common perceptual fate to the entire collection of micro-elements entering into these gestalts, something seems to be acting to hold them together as a group which the system recognizes as such. In a computer system this would be easy. For example once it were decided what pixels belong to an edge, on could simply put these points into linked list, approximately in their order along the edge. But how is a grouping with similar effect to be encoded by a distributed tissue of neurons, which need to act in a massively parallel way to achieve reasonable speed in carrying out image operations, and to which artificial techniques like the use of long linked lists seem entirely foreign?
The present paper will attempt to explore this fundamental question. A firs step in doing so is to gather evidence which tell us what collections of point edges seem to exhibit signs of grouping into an edge gestalt, in that some element of common fate can be seen to attach to them. In this attempt, we shall make use of the following figure, and others like it, described below. The figure contains two regions filled with coherently moving random dots, the outer moving downward, the inner circular region moving to the right. patient viewing of this image will show that it gives rise to two quite different percepts, both reasonably stable. (Since, depending on the speed of your computer and display screen, Browsers may have difficulty in displaying the successive frames of Figure 1 and our other animation at adequate speed, so all figures given as animated gifs re also given as Shockwave animations, which generally display more rapidly. If neither of these display methods achieves speed sufficient to see the effects discussed, you may need to use you browser (specifically, its "Save this link as..." capability) to download the animated gifs and display them in a specialized tool like 'GifBuilder', which can display frames at higher speed, and allows the display rate to be controlled.) In the first percept one sees a circle, seeming somehow to comprise the figure's foreground, having an illusory motion to the right. In the second percept one sees the outer random-dot frame moving downward as a 'foregound object, and the seems to see the circle as a background viewed through a transparent hole in the frame as this moves steadily downward. Once captured, each of these percepts is seen for a while, and then suddenly seems to switch to the other. It is always the whole figure that switches, never a part of it. (Note that the first of these two percepts seems to be the strongest. To help the alternative percept emerge, concentrate on some part of the frame that is not too close to the circle, until the frame starts moving coherently downward. Tricks like this are also useful in viewing our later figures.)
Figure 1. A figure distinguished by motion edges only.
We can attempt the following explanation for the two surprising percepts seen in Figure 1. The two regions seen in the figure are filled with different fields of coherent motion, one horizontal, the other vertical. Thus one will be a bright region in the population of cells sensitive to rightward horizontal motion ('horizontal motion image'), while the other will be bright in the 'vertical motion image'. The edge separating the two regions will be present in both, as and edge between a light and a dark region. In both these images the edge present will carry the corresponding motion signal, much as it might at the boundary between a coherently moving textured area and a dark background. Thus the horizontal motion image will generate a horizontally moving edge percept, and the vertical motion image will generate a vertically moving edge percept. One of these two will repress the other and thus 'capture' the edge, much as presumably happens in binocular fusion. If the edge is captured by the vertically moving percept, the region outside the circle will be seen as having an edge moving coherently downward and so will be perceived as an area with a circular hole in it moving downward. If the edge is captured by the horizontally moving percept, the circle will have an edge moving horizontally and so will be perceived as an object moving horizontally.
Given our interest in examining the coherence of such fundamental gestalts as edges, it is reasonable to separate the two halves of the moving circle seen in figure 1 by an obscuring bar, and then to ask whether the two half edges into which this bar divides the outer edge of the circle seen in can be differently captured, presumably leading to the perception of a downward moving surround in one half of the figure and a rightward-moving semicircle in the other. This is done in the following figure. Different capture of the two halves of the figure does not seem to occur.
Figure 2. A circle distinguished by motion edges only, but partly obscured by a bar.
Figure 3. A rectangle and annulus distinguished by motion edges only.
Figure 4. A rectangle and annulus distinguished by coherent and chaotic motion edges.
Figure 5. A figure involving three distinct coherent motions.
The motion after-effect and its use as a tool for investigating motion-sensitive cell populations.
Figure 15. A coherent motion which produces a motion after-effect
A minimal set of cell populations capable of sensing all directions of coherent motion might consist of as few as two populations, if we are willing to assume that they are constitutively on in the absence of the motion to which they are sensitive, and that notion in the opposite direction lowers their activity. But this would imply that the cells grew less fatigued in the presence of this opposed motion, which would therefore be a direction producing no motion after effect, a phenomenon for which there is no evidence.
If we therfore abandon this possibility and think if cell populations sensitive to motion in one direction and insensitive to the opposite motin, a minimal set could const of three populations perferably sensigtive to ,