Abstract: PURPOSE: The visual system continuously adapts to the statistical properties of the environment. Nonetheless, the mechanisms underlying the learning of naturalistic images, which contain the richest statistical dependencies, remain unclear. Here we utilized a computer vision approach to parameterize the co-occurrence statistics in naturalistic textures, and investigated the behavioral characteristics of learning these statistics.
METHODS: We utilized a computational model (Portilla & Simoncelli, 2000) to capture the statistics embedded in naturalistic textures and to synthesize textures that are perceptually indistinguishable from the original ones. Subjects underwent training over weeks to discriminate naturalistic texture from its spectrally-matched noise, which only differed in their high-order statistics. Experiment 1 evaluated the contribution of different types of statistics to learning, including linear statistics that reflect the periodicity and global structure, energy statistics that depict structures like edges and corners, and phase statistics that represent luminance gradients from shading. Experiment 2 tested the statistical specificity and location specificity of learning, which is crucial in determining the neural locus of perceptual learning. Experiment 3 further explored the location specificity by comparing the full learning courses between the transfer and the original training.
RESULTS: 1) High-order statistics played a critical role in naturalistic texture learning. 2) The learning effect was specific to high-order statistics and retinal location. 3) An accelerated learning was found at un untrained location.
CONCLUSIONS: By manipulating co-occurrence statistics in naturalistic textures, the present study built a link between perception and statistical learning. Our findings indicate a multi-stage statistical learning that bridges the gap in the learning-induced plasticity between the early to mid-level visual system.

Key words: naturalistic texture, statistical learning, visual plasticity, psychophysics