First-order and second-order visual processes in the perception of motion and tilt.

It has been known for many years that the mammalian visual system contains neurones which respond selectively to the orientation of contours in the image, and to the direction of their movement. Numerous perceptual studies have also found evidence for the existence of orientation and direction specific channels in the human visual system. A new theoretical framework is currently emerging, which sub-divides direction selective channels into first-order and second-order classes. Data from two experiments reported here show that each class of mechanism can be activated in isolation from the other, using appropriate motion stimuli. Both can mediate direction discrimination, and both give rise to after-effects. Further, analogous illusions of tilt are presented which demonstrate that the first vs second-order distinction applies to orientation coding processes as well. The dichotomy between first and second-order processes therefore may reflect a more general property of the visual system’s organisation. A recent evaluation of psychophysical data on motion perception proposed that motion detectors in the human visual system can be divided into two broad classes: first-order detectors, and second-order detectors (Cavanagh & Mather, 1989). Broadly, firstorder detectors respond to the movement of contours defined by intensity gradients in the image (i.e. first-order differences in the intensity distribution), whereas second-order detectors respond to the movement of contours defined in more abstract terms (texture density, flicker, binocular disparity, motion parallax, etc.). Chubb and Sperling (1988) have developed a range of second-order stimuli, which are invisible to first-order detectors (“drift-balanced”). The two classes of detector may correspond to, or at least overlap with the two processes previously identified as “short-range” and “long-range” (Braddick, 1980). Figure 1 illustrates motion stimuli designed to activate each class in isolation from the other. The leftmost panel is an xt plot of a standard apparent motion stimulus-space (x-position) is represented on the horizontal axis, and time on the vertical axis. A dark bar is shown occupying a series of static positions at successive time intervals. If the spatial and temporal steps are chosen appropriately, an illusion of rightward apparent motion will be seen by an observer. One can represent a simple first-order detector which responds to this rightward motion as having a receptive field in the xf plot with the shape shown in the figure, so that it is tuned to a particular “spatiotemporal orientation”. The tilt of the receptive field’s long axis specifies its preferred velocity (distance travelled per unit of time). Detectors tuned to other velocities, either rightward or leftward, would have receptive fields with different spatiotemporal orientations. There is good evidence that the visual system does possess motion detectors tuned to spatiotemporal orientation, and several models have been proposed (Adelson & Bergen, 1985; Ross & Morrone, 1986; Emerson, Citron, Vaughn & Klein, 1987; Watson & Ahumada, 1985). Consider the middle panel of Fig. 1. Here, a bar is shown alternating between two positions over successive time intervals. Contrast polarity is maintained for rightward shifts, but reverses for leftwards shifts. It was first reported by Anstis in 1970 that shifts accompanied by contrast reversals produce reversals in apparent direction, and recent perceptual, computational and physiological data confirm that directional signals are simply inverted by contrast reversal