Abstract
Objective: We hypothesize a cascade model of cortical processing in which the same canonical computations (linear filtering, F, rectification, R, and normalization, N) are repeated across a hierarchy of stages (F1R1N1, F2R2N2, etc). We tested for the existence and orientation selectivity of 2nd-stage normalization (N2).
Methods: Stimuli were either contrast-modulated (CM) or orientation-modulated (OM) 2nd-order stimuli with bandpass noise carriers and low-frequency sinusoidal modulators. Observers performed a 2AFC task to identify which segments of an annular target region contained 2nd-order modulation. The target was either presented alone or embedded in a surround (inside and outside of the annulus). Three types of surround stimuli were used: no 2nd-order modulation, or 100% modulation amplitude in which the modulator orientation was either parallel or orthogonal to the target modulation. For each condition, we estimated the 2nd-order modulation depth for the target corresponding to discrimination threshold.
Results: Four out of four subjects (for CM stimuli) and four out of five subjects (for OM stimuli) showed significant threshold elevation in the full-modulation-surround conditions compared to the zero-modulation-surround condition, indicative of normalization. For CM stimuli, all subjects showed an orientation-specific effect, exhibiting higher thresholds for parallel than for orthogonal surrounds. For OM stimuli, 2nd-order surround suppression was not consistently orientation-selective. In no case was there evidence of significant 1st-order surround suppression (zero-modulation vs. target-alone conditions).
Conclusion: 2nd-order spatial coding uses similar computations to 1st-order, including normalization (surround suppression). For some forms of 2nd-order modulation, normalization is orientation-selective.
NIH grants R01-EY016752 (DJH) and R01-EY016165 (MSL & DJH).