Bayesian inference of form and shape

Abstract

The ability to visually perceive two-dimensional (2D) form and three-dimensional (3D) shape is one of our most fundamental faculties. This ability relies on considerable prior knowledge about the way edge elements in an image are likely to be connected together into a contour as well as the way these 2D contours relate to 3D shapes. The interaction of prior knowledge with image information is well modeled within a Bayesian framework. We review here the experimental evidence of shape perception seen as a Bayesian inference problem.