A central challenge of vision research is to understand how the brain translates patterns of light on the retina into useful information about the shape of objects. Shape is a more reliable indicator of object identity than color, texture, or any other visual property, yet the representation of shape in the visual system is still poorly understood. Theories from visual psychology have suggested that the visual system represents objects in terms of geometrically-defined parts, or in terms of two-dimensional image features or fragments. Which theory best describes the representation of shape in the visual system? The debate in psychology has revolved around defining the representation of shape, but neurophysiology and imaging studies have revealed a hierarchy of regions in the visual system. From this perspective, the relevant question is not whether the visual representation is shape-based or view-based, but which features or dimensions are represented in which areas. Some progress has been made on this question, but many studies of shape representation use stimuli that only vary in one or two dimensions. Constrained stimulus variation makes it impossible to know whether a neuron (or fMRI voxel) might represent other dimensions that do not vary in a particular stimulus set. In order to better describe what shape features or dimensions the visual system represents, it is necessary to use stimuli that vary in multiple dimensions. Natural scenes can be used as stimuli, but the shape variation in natural scenes is difficult to parameterize. Here, we propose to create a widely varied stimulus set using Blender, a 3D graphics and animation program. The stimuli will consist of realistic rendered objects, and will vary in many dimensions (e.g. boundary curvature, part arrangement, and pose). We will use meta-data from Blender to parameterize these shape features for each stimulus image. We will use the Gallant Laboratory's Non-Linear System Identification (NLSI) framework to discover how these shape parameters are reflected in BOLD fMRI brain responses in different regions throughout the visual system. The proposed work will thus provide a critical test of the generality of past work on shape representation, as well as a more complete understanding of which shape dimensions are represented in which visual areas. This proposal is in keeping with the NIH's mission to understand normal visual function. A deeper understanding of shape representation across the visual hierarchy will facilitate the development of visual prosthetics, and insight into visual processing can provide insights into information integration in other senses.
The proposed project is in keeping with NIH's mission to understand the normal function of the visual cortex. Activity in shape-selective visual areas correlates with subjective perceptions and causally affects behavior;thus the study of shape representation is truly the study of visual consciousness. A deeper understanding of shape representation across the visual system will contribute to the development of visual prosthetics, serve as a model for information integration in other senses, and potentially set the stage for a more thorough understanding of individual differences in perception and attention.