Abstract
The steady visual evoked potential (SSVEP) is a measurement that is dependent on the shared activity of an ensemble of neural responses in the visual pathway. It provides a non-invasive technique that can theoretically provide important insights into the processing of visual stimuli. Here, we investigate the information that SSVEP can provide regarding the identity of individual natural scenes. We took a geometric state-space approach to this question by modeling the transformation between the state-space of human evoked potentials and images drawn from different locations within a natural scene state-space. Data were gathered in a steady-state visual evoked potential paradigm whereby participants viewed grayscale visual scenes while undergoing 128-channel EEG. Scene images were contrast modulated at a sinusoidal flicker rate of 5Hz for 6000msec while participants engaged in a distractor task at fixation. Electrode data with the highest signal-to-noise ratio were submitted to a principal component (PC) analysis on a participant-by-participant basis. Stimuli in image state-space were mapped to their response location in PC-defined neural state-space with an averaged accuracy of 73% (explainable variance accounted) using a simple Fourier filter-power model that far exceeded models based on pixel statistics. Interestingly, spatial frequencies at and above 4cpd yielded the best mapping, suggesting that images are mapped in early neural state-space according to a simple whitening process.