Central Processing of Visual Information

Weill Medical Coll Of Cornell Univ, New York NY

Investigators

Linked publications & trials

Abstract

DESCRIPTION (provided by applicant): The overall goal of the research is to determine the nature of early vision's representation of the visual world, and the design principles of visual cortical function. The proposal centers on testing a hypothesis that embodies the efficient coding principle: that the representation of the visual world is based on extracting the local image statistics that are the most informative about natural images. That is, in order to make optimal use of its limited resources, the visual system omits an explicit representation of aspects of natural images that are predictable, so that it can concentrate its resources on what is not predictable, and therefore, informative. While this notion has been widely successful as an organizing principle for understanding retinal processing, its application to visual cortex has been much more limited, in part because of the complex structure of natural scenes. The planned approach overcomes this barrier, through the use of a novel mathematical strategy that recognizes the complex characteristics of the sensory environment. Implementing this approach leads to a range of specific (and occasionally surprising) predictions, owing in large part to the complex structure of natural scenes and their statistical regularities. If supported by the experimental results, the resulting view of early visual processing will broaden current notions of the calculations that neural circuitry performs, beyond simple filtering and feature extraction. Aim 1 consists of three sets of experiments to test predictions of the hypothesis in a comprehensive way. Specifically, Aim 1A determines whether the distribution of local image statistics accounts for perceptual sensitivity to first, second, third, and fourth-order statistics. Aim 1B determines whether the scale-invariance of image statistics in natural scenes is mirrored by a scale-invariance of perceptual salience. Aim 1C extends the analysis from binary images (used in Aims 1A and 1B for simplicity) to gray-level images, and uses psychophysical results to predict as-yet unrecognized regularities in natural images. Since it is important to understand not only the overall goal of visual processing, but also how it is achieved, Aim 2 characterizes the computational mechanisms that underlie extraction of image statistics. Specifically, Aim 2A focuses on dynamics, and how these calculations evolve as the perception of an image advances in time from a gist to a detailed appreciation. Aim 2B determines whether image statistics are calculated via a discrete set of mechanisms likely to be embodied in dedicated circuitry, vs. a continuum of virtual ones. Successful completion of this research will advance the understanding of the goal and design principles of cortical visual processing, and thus, will support the rational design of advanced therapeutic modalities, such as neural prosthetics for loss of visual function.