Local spatio-temporal analysis in vision systems

Abstract : The aims of this project are to (1) develop a physiologically and psychophysically based model of low-level human visual processing (a key component of which are local frequency coding mechanisms), (2) develop image models and image-processing methods based upon local frequency coding, (3) develop algorithms for performing certain complex visual tasks based upon local frequency representations, (4) develop models of human performance in certain complex tasks based upon our understanding of low-level processing, (5) develop a computational testbed for implementing, evaluating and visualizing the proposed models and algorithms, using a massively parallel computer. Progress has been substantial on all aims. The highlights include, (1) completion of a number of psychophysical and physiological experiments revealing new, systematic and exciting properties of the primate (human and monkey) visual system, (2) further development of image models that can accurately represent the local frequency structure in complex images, (3) near completion in the construction of the Texas Active Vision Testbed, (4) development and testing of several new computer vision algorithms dealing with shape-from-texture, shape-from-stereo, and depth-from-focus, (5) implementation and evaluation of several new models of human visual performance, (6) evaluation, purchase and installation of a MasPar parallel computer.