Visual Filling in and Completion

Why Filling in and Completion Are Necessary

The need for boundary completion can be understood when considering the following example. If one stands under a tree and looks up while holding a pencil in front of one's eyes, the tree's branches and pencil will merge into a single complex distribution of dark and light on the retina. On the retina, the branches are “attached” to the pencil, but clearly, this is not what is perceived. Somehow the visual system “knows” that the branches of the tree continue behind the pencil in front, despite the fact that the continuation of the tree branches is not present in the retinal input. How the brain determines that the two parts of a tree branch occluded by a pencil (or another object in front) belong together is the problem of boundary perception.

How Filling in and Completion Work

Introducing a few basic concepts by means of metaphor will fully elucidate the problem of boundary perception, and will also give insight into the problem of surface perception. The visual system can be compared to a camera (the “retina”) in which tiny light sensors each send an electrical signal to a pixel on a monitor (e.g., area V1, the primary visual cortex). This metaphor is based on the finding that there is a roughly point-by-point anatomical connection between locations on the retina and sites of electrical activity in V1 (V1 is therefore called a retinotopic map). However, to complicate matters, there is only a small minority of neurons in V1 that simply represent light values [Page 1076](or brightness) in specific retinal locations. Instead, most neurons in V1 represent local contrast in the image. Using the monitor metaphor, most pixels in the monitor are connected to two neighboring light sensors in the camera, rather than just one. These pixels are only turned on if two neighboring light sensors in the camera signal different light values, and are not turned on when the neighboring light sensors signal the same light value. This circuitry in the visual system emphasizes contrasts (often related to boundaries), while most information about surfaces is lost; hence the additional problem of surface perception. The primary role of boundaries in object recognition is revealed by the ease with which we interpret line drawings and cartoons, in which recognition is based only on boundaries and not on the filling in of surfaces.

Figure 1 Illustrations of Boundary Completion and Surface Filling-In

Sources: Investigators who introduced these figures to experimental research are Kanizsa (Figures 1a and b), and Craik, O'Brien, and Cornsweet (Figures 1c and d).

The problem of boundary and surface perception can now be summarized by asking the questions how do individual neurons in the visual system cooperate to link local information that belongs to the same boundary, and how do they cooperate to reconstruct the percept of surfaces that seems to be lost as a price for the ability to see local contrasts. In essence, the solution to these problems resides in an exchange of information between neighboring neurons (through anatomical connectivity that goes beyond the scope of this entry).

...