Abstract
The human visual system can detect not only modulations in luminance (first-order stimuli), but also modulations of these modulations (second-order stimuli). For example modulations in the contrast, orientation, or motion of a first-order “carrier” stimulus. Second-order modulations are substantially more difficult to detect than equivalent first-order modulations. This difference would be explainable if we understood the processing mechanisms of second-order stimuli, however these are not yet clear. In order to compare the different types of second-order processing and allow comparisons to be made against first-order, we used the pedestal masking method to measure dipper functions for three types of second-order stimuli: contrast-, orientation- and motion-modulated. The contrast-modulated stimuli are constructed by modulating a 45°, 4 c/d sinusoid grating carrier with a horizontal sinusoid grating envelope with a spatial frequency of 0.5 c/d. The orientation-modulation stimuli are made by adding two contrast-modulated stimuli with perpendicular carrier gratings and opposite envelope phases together. The motion-modulation stimuli are made by adding two contrast-modulated stimuli with opposite envelope phases but the same orientation drifting in perpendicular directions. The data are fit using maximum likelihood with a modified version of the Legge & Foley (1980) contrast response function. We find the dipper shapes similar to first order (same exponents) for the contrast, orientation and motion stimuli. Compared to contrast-modulation (from which the other stimuli are constructed) both the orientation- and motion-modulation conditions show an increased saturation constant (8 and 2 times higher respectively) consistent with increased divisive suppression. The motion-modulation condition also has an increased internal noise about 1.2 times that for contrast-modulation. In comparison with first-order results from previous studies we find all three second-order conditions have increased internal noise and greatly increased saturation constants. We consider preliminary designs for model architectures that may account for our results.
Meeting abstract presented at VSS 2015