Contour Processing in primate visual cortex

The interpretion of two-dimensional images in terms of a three-dimensional world is a basic task of vision. The human visual system performs this task with great ease, so that we hardly become aware of it. Indeed, we see the world three-dimensional and if we did not know about the eye’s optics and retinal images, we would perhaps never suspect that our vision is based on flat images. Vision decomposes a scene into objects as if we would touch and handle things in space. On the other hand, the technical difficulties of interpreting images are well known. In general, objects occlude one another and foreground and background structures are cluttered up in the image. To discriminate objects, one has to find the occluding contours first, but this proved to be difficult when the objects are not known. Contours indeed seem to play a fundamental role in human vision. We can recognize many objects just by their contours, and if we look around, most objects appear to be bounded by clear contours. However, in the corresponding array of gray values the contours may not be clear at all. The photograph in Fig.1 is meant to demonstrate this (since we cannot turn off the processing in our brain we have to exaggerate the problems in order to see them). We perceive contours even at sites where there is no discontinuity in the image. Fig.2 shows more examples of such “anomalous” contours. Does this perception depend on the knowledge about the possible shapes of objects? Is the definition of contours really prior to recognition? Perhaps we perceive the contours only because (and after) our brain infers the objects. This wouid be the “cognitive” interpretation.