Abstract
The output of linear mechanisms comprising the “first-order”, or Fourier mechanisms is known to be followed by at least two nonlinearities and by higher-level circuits that sum outputs of first-level mechanisms with very specific tuning characteristics, to form specialized edge and texture detectors (Olzak & Thomas, 1999). On the other hand, the system(s) thought to process static second-order, or non-Fourier stimuli, are already specified as linear-nonlinear-linear cascades. They are known to be tuned with respect to spatial frequency and orientation, albeit with some differences from the first order system. The question we ask here is what happens to the output of second-order mechanisms with very different orientation tuning. We created second-order sinusoidal patterns by contrast-modulating binary random noise. In each condition of three different experiments, observers discriminated between two very similar second-order patterns on the basis of small differences in orientation, spatial frequency, or contrast in the second-order modulation. In two separate control conditions, stimuli were simple sinusoids orientated either vertically or horizontally. Potential nonlinear processes were isolated in two masking conditions, which added a non-varying second-order mask of orthogonal orientation to the control patterns. Potential combining mechanisms were tested for in configural-effect conditions. In these, cues to discrimination were present in both the vertical and horizontal components of plaid second-order patterns. They were combined in two different ways. Any difference in performance in the two conditions signals a combining of information. The nature of the configural effect signals whether the combination is additive or differencing. Our results show significant masking by an orthogonal component, possibly indicating the presence of a widely-inclusive gain control pool, and a configural effect that suggests higher-level summing circuits. We conclude that the linear-nonlinear-linear process is iterative within the visual processing system.
This work was suppored by NIH Grant EY13953 to LAO