Outline of a Computational Theory of Human Vision

Abstract. Human vision is a powerful yet highly efficient processing system. Drawing on an extensive review of empirical findings and theoretical foundations of human visual-spatial perception originating from various disciplines, results of an in-depth analysis of the human visual perception process will be presented. Several stages in the visual pathway will be identified that segment into processes in eye and retina, control processes for exploration and sampling, processes in the primal visual cortex, feature detection, and, furthermore, object recognition processes. Based on the review, a computational theory of human vision will be drafted with emphasis on developing an architectural model. The model introduced will couple enhanced perceptrons over an intermediated layer, which is responsible for controlling exploration and sampling, to primal feature decomposition modules. A feature based router then is responsible for distributing this preprocessed input to concept detector components. The architecture will allow both bottom-up and top-down data flows. Moreover, it will facilitate ’lazy ’ processing by introducing means for focused, concept driven attention.