Statistical characterization of tactile scenes in three-dimensional environments reveals filter properties of somatosensory cortical neurons

Abstract

Natural scenes statistics have been studied extensively using collections of natural images and sound recordings. These studies have yielded important insights about how the brain might exploit regularities and redundancies in visual and auditory stimuli. In contrast, natural scenes for somatosensation have remained largely unexplored. Here we use three-dimensional scans of natural and human-made objects to quantify natural scene statistics at the scale of the human fingertip. Using measurements of distance, slope, and curvature from the object surfaces, we show that the first order statistics follow similar trends as have been observed for images of natural and human-made environments. In addition, independent component analysis of curvature measurements reveals Gabor-like basis vectors similar to those found in natural images. A simple neural model using these filters showed responses that accurately capture the statistics of responses in primate primary somatosensory cortex.