Second - Order Reversed

There is mounting evidence that there are three parallel streams of visual motion computation—firstand second-order systems that are primarily monocular and a third-order binocular system (Lu & Sperling, 1995a, 1995b, 1996c). The first-order system extracts motion from drifting luminance modulations, and the secondorder system extracts motion from drifting texture contrast modulations. These primarily monocular systems use motion energy analyses (Adelson & Bergen, 1985) or, equivalently, elaborated Reichardt detectors (van Santen & Sperling, 1984, 1985). Both primarily monocular systems are fast (temporal cutoff frequency at 10–12 Hz) and approximately equally sensitive to a wide range of spatial frequencies (0.6–4.8 cpd). These relations are illustrated in Figure 1. The third-order system is binocular, in the sense that it is indifferent to the eye or eyes of origin of successive images. When successive images alternate between eyes, so that motion extraction requires the combination of left-eye and right-eye information, only the third-order system is effective. Lu and Sperling (1995a) proposed that the third-order system derives its input from a dynamic salience map that records the moment-to-moment positions of the most salient stimulus features (i.e., figure vs. ground). The motion of areas marked as figure is computed from this dynamic map, just as the movement of areas marked by more (or fewer) photons is computed by the first-order motion system and the movement of areas marked by more (or fewer) features is computed by the second-order system. The third-order motion system is relatively slower (temporal cutoff frequency is 3–4 Hz), has lower spatial resolution than the firstor second-order system (Lu & Sperling, 1995b), and can be influenced by attentional instructions (Lu & Sperling, 1995a). Nishida (1993) first observed second-order reversed phi. Here, we describe novel demonstrations of second-