Abstract
Capitalizing on predictability in the environment is a fundamental principle of biological sensory systems. To understand sensory encoding, it is therefore essential to have a comprehensive description of the regularities in one’s experience of the natural world. Statistical descriptions of natural images have delineated key non-uniformities in the relative probabilities of basic visual features, such as contrast, spatial scale, contour orientation, and depth. The information encoded in visual cortex, however, is not a direct copy of the light falling on photoreceptors. We wanted to determine how these well-known regularities in natural images are altered by the processing of early visual pathways. Using a large image dataset, we simulated the transformations these pathways impose on the input to cortex. The simulations included the receptive fields and non-linear response properties of magnocellular (M) and parvocellular (P) pathways, as well as ON and OFF pathway divisions. Statistical patterns of natural images were thus transformed into statistical patterns of afferent neural signals. This analysis indicates that there are regularities in these signals beyond those that have previously been appreciated from the direct analysis of natural images. Not surprisingly, activity patterns diverged with regards to spatial scale between the M and P pathways. However, within the M and P pathways, the ON and OFF divisions also differed substantially in the amount of activity devoted to different levels of contrast and to different spatial scales. We propose that these asymmetries between ON and OFF signals reflect an additional layer of predictability in the visual input that has been exploited when encoding basic visual features in cortex. These ON/OFF pathway differences also suggest that some previously measured response asymmetries to bright and dark stimuli in cortex may have a pre-cortical origin.
Meeting abstract presented at VSS 2015