Abstract
Efficient processing of sensory data requires adapting the neuronal encoding strategy to the statistics of natural stimuli. One area where this concept has been applied is the perceptual saliency of visual textures: multiple works have shown that humans are most sensitive to textures containing multipoint correlations that vary the most across natural images. The neuronal mechanisms underlying such adaptation are not understood, but investigating them in depth requires invasive experiments that are impossible in humans. Therefore, it is important to test this phenomenon in animal species that allow for such experimental manipulations. Recently we have shown that rats can be trained to discriminate between binary textures containing unstructured random noise and textures defined by single- to four-point correlations among nearby pixels. We observed a sharp decrease in sensitivity from 2- to 4-point correlations and a further decrease from 4- to 3-point. This ranking fully reproduces the trend previously observed in humans and in the variability of multipoint correlations in natural images. Building on this result, we started performing extracellular recordings of neural activity from primary visual cortex (V1) and extrastriate areas of anaesthetized rats, passively viewing textures containing multi-point correlations or unstructured noise. We found that, in V1, the highest fraction of selective units (p-value<0.05, ANOVA) were tuned for 1-point correlations, followed by 2-,4- and 3-point. In extrastriate cortex, we found the largest fraction of units to be selective for 1-point, followed by 4-, 2- and 3-point correlations. Finally, we confirmed a difference in neural representations of texture statistics between striate and extrastriate cortex with a population decoding analysis. These results explore the neuronal underpinnings of a classic instance of efficient coding that, to date, has been investigated mainly at the perceptual level, and provide an early indication of the relative contribution to this process by distinct visual cortical areas.