Neural Models of Motion Perception

Apparent motion (AM) involves two or more “frames” of images, where elements are displaced from frame to frame to elicit the percept of motion. An important question is how a certain element, or feature, is matched across frames to yield the veridical motion percept without false target localization. This is known as the “correspondence problem” (Julesz, 1968; Ullman, 1979). One possibility is that bottom-up, hard-wired neural mechanisms compute motion automatically, without solving explicitly the correspondence problem. At the other extreme top-down motion-tuned mechanisms can first extract specific features of moving objects (such as edges, corners, etc.) and then track these features from one frame to the next to achieve a solution to the correspondence problem. It is quite possible that a wide variety of motion sensing units exist in the visual system, covering the entire spectrum that is defined between the two extremes outlined above. Neural mechanisms that are tuned to a specific direction of motion have been found in the striate cortex, as well as in the extrastriate areas, such as area V5, otherwise known as MT (middle temporal), and area MST (medial superior temporal) (Tovee, 1996).