An Introduction to Perceptual Organization

Perceptual organization refers to the ability to impose organization on sensory data, so as to group sensory primitives arising from a common underlying cause. The existence of this sort of organization in human perception, including vision, was emphasized by Gestalt psychologists. Since then, perceptual organization ideas have been found to be extremely useful in computer vision, impacting object recognition at a fundamental level by significantly reducing the complexity. In this paper, we provide a short introduction to perceptual organization, outlining its genesis in human vision, followed by a Bayesian interpretation of the organization process. This Bayesian framework lets us systematically design organizational processes suitable for each application. We also provide a unifying framework for the various spectral partitioning frameworks, such as average cut and normalized cut, which has been found to be very effective in computer vision. The impact of perceptual organization is showcased using examples of target segmentation for ATR and motion segmentation.