Abstract
Recent psychophysical and neurophysiological evidence shows that primate early visual cortex computes statistical structure of natural images, which may underlie rapid and efficient perception and recognition of complex scenes and surfaces. In order to understand the temporal dynamics of neural encoding of these statistical information, the present study analyzed the correlation between visual evoked potentials (VEPs) to natural texture images and simple image statistics. In our procedure, 8 observers viewed 111 achromatic images of various natural textures (mean luminance of 54 cd/m2) each of which was successively presented for 500 ms in random order. Each image was repeatedly shown for 30 times in separate blocks. During the experiment, EEG signals were recoded from 19 electrodes. Following preprocessing of EEG signals, the average VEPs for each image was calculated across repeated presentations and observers. We then computed single correlation coefficient between occipital VEP amplitudes for 111 images with image statistics as used in Portilla-Simoncelli texture analysis, which has been suggested to be encoded in V1 and V2. The analysis revealed dynamic changes in correlated image statistics. For example, VEPs at 250, 300, and 350 ms showed high correlations with SDs and kurtosis at low, mid, and high spatial frequencies, respectively. Particular periods (250–400 ms) of VEPs were also correlated with higher-order image statistics such as cross-orientation and cross-frequency linear/energy correlations. These results seem to suggest that the early visual cortex encodes image statistics in textures with different temporal dynamics.
Acknowledgement: The Commissioned Research of National Institute of Information and Communications Technology, JSPS KAKENHI JP15H05916, JSPS KAKENHI JP18H04935