Abstract
Normalization models of orientation masking define the percept of the target stimulus as the linear sum of the target and mask followed by rectification (i.e., a non-linearity). However, recent evidence (e.g., Baker & Wade, Cereb. Cortex, 27, 254-264) has suggested that stimulus combination in the early visual system is non-linear: target and mask signals undergo rectification prior to combination. Here, we aim to define the characteristics of target and mask combination (linear vs non-linear) when they have the same or different orientation (i.e., cross-orientation suppression). We used Steady-State Visual Evoked Potentials (SSVEPs) to record neural responses at flicker frequencies of the target (f1) and mask (f2), in addition to the characteristic signatures of non-linear response components (i.e., inter-modulation terms: f1+f2 and f2-f1). Target stimuli were 2° horizontal sinusoidal gratings with spatial frequency of 4 cycles/°, flickering at 5Hz, and generated at 5 different contrast levels (14, 20, 26, 32 and 36dB). Masks were identical to the target in size and spatial frequency, but flickered at 7.5Hz, were generated at a fixed contrast of 32dB, and offset in orientation to the target by 0° to 90° in steps of 15°. Response amplitude at f1 showed evidence of masking: amplitude increased monotonically with contrast and was rightward shifted compared to baseline for all but the co-oriented mask. Intermodulation term amplitude increased with target contrast when the target and mask were co-oriented or offset by 15°, but reduced to noise levels at larger mask orientation offsets. We implemented 5 model variants that define stimulus combination from fully linear to fully non-linear, and verify which is most apt at generating predictions that match our SSVEP data. We find our effects to be best explained by a fully non-linear model, and implement a geometric analysis to define the response surface of non-linearities measured with SSVEPs.
Meeting abstract presented at VSS 2018