Purchase this article with an account.
Taiki Orima, Isamu Motoyoshi; Analysis and systhesis of natural texture perception by EEG. Journal of Vision 2020;20(11):648. doi: https://doi.org/10.1167/jov.20.11.648.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Recent psychophysical and neurophysiological evidence shows that human early visual cortex encodes the image statistics of natural textures. To investigate how the neural representation of texture statistics evolves over time, we performed a reverse-correlation analysis between visual evoked potentials (VEPs) to natural texture images and image statistics computed from those same textures. 15 observers viewed 166 achromatic images of various natural textures presented foveally for 500 ms in random order (x 24 for each image), and EEG signals were recoded from 19 electrodes. Additional recordings were made for the Portilla-Simoncelli (PS)-synthesized and phase-randomized versions of the original natural textures. We calculated correlation coefficients between VEPs and image statistics, including subband moment statistics and cross-subband correlations. Our analysis showed that, for all stimulus versions, image statistics are manifestly correlated (max r=~0.8) with texture VEPs at systematically different timings; e.g., subband SDs at low, middle, and high spatial frequencies were correlated with VEPs at 100, 120, 150 ms respectively. Significant correlations were also observed between VEPs and PS statistics. Based on these data, we further carried out a VEPs-to-PS statistics inverse reconstruction by applying the partial least squared (PLS) regression analysis between VEPs (125 points) and compact PS statistics (110 vectors). 5 components were obtained, as determined by 10 folds cross-validation. Coefficient matrixes generated from training data (90 % of stimuli) successfully synthesized a portion of textures that were perceptually very similar to the original PS-synthesized textures. These findings suggest that simple VEPs elicited by natural textures contain information about the dynamics of cortical responses to image statistics that is sufficiently rich to predict human perception.
This PDF is available to Subscribers Only