Abstract
The environments in which an organism lives and the tasks it performs within those environments shape its perceptual systems through evolution and experience. This is an obvious statement, but it implies several fundamental components of research that are needed if we are going to gain a deep understanding of perceptual systems. The first is to identify the natural tasks and sub-tasks that are performed by the organism under natural conditions. The second is to measure and analyze those specific environmental properties (natural scene statistics) relevant for performing the tasks. The third is a computational analysis to determine how a rational (ideal) perceptual system would exploit the measured environmental properties to perform the tasks. This component is critical because it provides insight into the information contained in the natural stimuli and it can suggest principled hypotheses for the neural mechanisms the organism might use to exploit that information. The fourth component is to formulate specific hypotheses for neural mechanisms, based on the first three components, and test them in physiological and behavioral studies that capture the essence of the natural task. This general approach is illustrated with a study of contour grouping that combines measurements of natural scene statistics, derivation of ideal Bayesian observers that exploit those statistics, and psychophysical experiments that compare human and ideal performance. This study and other recent studies demonstrate the great potential of “natural systems analysis” for producing advances in behavioral science and systems neuroscience.
GeislerW. S.PerryJ. S.IngA. D. (2008) Natural systems analysis. In: RogowitzB.PappasT. (Eds.), Human Vision and Electronic Imaging, SPIE Proceedings, Vol. 6806.
GeislerW. S.PerryJ. S. Contour statistics in natural images: Grouping across occlusions. Visual Neuroscience.
Supported by NIH grant EY1147.