Linking Dynamical Perceptual Decisions at Different Levels of Description in Motion Pattern Formation: Psychophysics

Article excerpt

The relationship between local-level motion detection and higher level pattern-forming mechanisms was investigated with the motion quartet, a bistable stimulus for which either horizontal or vertical motion patterns are perceived. Local-level perturbations in luminance contrast affected the stability of the perceived patterns and, thereby, the size of the pattern-level hysteresis obtained by gradually changing the motion quartet's aspect ratio. Briefly eliminating luminance contrast (so nonmotion was perceived during the perturbation) eliminated pattern-level hysteresis, and briefly increasing luminance contrast (so motion was perceived during the perturbation) increased pattern-level hysteresis. Partially reducing luminance contrast resulted in bistability during the perturbation; pattern-level hysteresis was maintained when motion was perceived, and eliminated when nonmotion was perceived. The results were attributed to local motion/nonmotion perceptual decisions in area Vl affecting the magnitude of the activation feeding forward to motion detectors in area MT, where the stability of pattern-level perceptual decisions is determined by activation-dependent, future-shaping interactions that inhibit soon-to-be-stimulated detectors responsive to competing motion directions.

An ongoing issue in our understanding of visual pattern formation concerns the relationship between different levels of description for the pattern. Although the traditional idea is that the relationship is hierarchical, with larger units built from smaller ones (see, e.g., Biederman, 1987; Cutting, 1986), it also has been argued that processing proceeds in the reverse, top-down direction (Hochstein & Ahissar, 2002; Sanocki, 1993). Upstream areas of the brain have larger receptive fields and respond to more abstract, global properties than do downstream areas (Maunsell & Newsome, 1987; Vogels & Orban, 1996), so different levels of description are likely to have their counterparts in different levels of neural processing.

The present research investigates the feedforward flow of activation from the detection of local element motions to the formation of simple global motion patterns. Psychophysical results in this article and computational simulations of these results in the accompanying article (Nichols, Hock, & Schöner, 2006) indicate that (1) pattern-level hysteresis depends on activation-dependent detector interactions and, therefore, on the magnitude of the activation that feeds forward from local to pattern levels, and (2) local-level dynamical decisions that result in motion or nonmotion being perceived (i.e., local Instability) affect pattern-level dynamical decisions by feeding forward different magnitudes of activation for the same stimulus. Dynamical decisions occur when the activation levels of stimulated detectors are attracted to and maintained near stable, fixed-point values (Hock, Schoner, & Giese, 2003), where each fixed point represents the joint activation state of all detectors relevant to the percept. It is assumed that the percept is embodied in those detectors whose activation is stabilized at values above the threshold level required for perception.

Hock, Kogan, and Espinoza (1997) and Hock, Nichols, and Espinoza (2004) have provided psychophysical evidence that detectors responsive to motion-independent spatial information compete with motion detectors in determining whether motion or nonmotion is perceived. They observed spontaneous switching between nonmotion and motion percepts, motion/nonmotion hysteresis, and the independent adaptation of perceived nonmotion and perceived motion. Our account therefore begins in area V1, where in addition to directionally selective motion detectors (Hubel & Wiesel, 1979) there are detectors responsive to motion-independent spatial information (Movshon, Thompson, & Tolhurst, 1978). Detectors in area MT are predominantly motion sensitive (Maunsell & Van Essen, 1983), so V1 is the more likely of the two sites for motion/nonmotion decisions. …