This award is under the International Postdoctoral Fellows Program, which enables U.S. scientists and engineers to conduct three to twelve months of research at foreign centers of proven excellence. The program's awards provide opportunities for joint research, and the use of unique or complementary facilities, expertise and experimental conditions abroad. This award will support a twelve-month postdoctoral research visit by Dr. Daniel L. Ruderman to work with Dr. David J. Tolhurst at the Physiological Laboratory at Cambridge. Their research proposes to apply natural scene statistics to basic problems in vision. Dr. Ruderman's graduate work measured the properties of images in nature. He found a very robust scaling in the statistics, meaning the character of natural scenes is invariant to scale. They now hope to answer the question "What does this scaling imply about the design of image processing systems in nature?" A long-held belief is that detecting edges is fundamental to image analysis. The optimal edge detector for biological vision is determined by the structure of edges in natural images. Similarly, the problem of constructing complete borders around objects can be phrased relative to the statistical properties of borders in nature. Do the optimal algorithms for edge and border finding need access to the complete image, or is local information enough? This is a crucial question for biology since neurons in the early stages of vision respond only to local image features. Does this fact limit the processing ability? Finally, optimal reconstruction of an image from noisy neural responses requires statistical knowledge of the images and the noise. Comparing this reconstruction ability with the performance of humans in psychophysical tasks represents a measure of inaccuracy in visual computations. Using natural scene statistics in answering these questions provides a novel framework for understanding vision in biology. The award recommendation provides funds to cover international travel and a stipend for twelve months.
