A Theoretical Framework for Visual Motion
Alan Yuille, Norberto Mauricio Grzywacz · The MIT Press eBooks · 1998
We propose a theoretical framework for visual motion. Models in this framework are constructed using two design principles: First, their mechanisms should be biologically plausible and consistent with experiments. Second, they should be well-engineered to perform well on motion data from realistic scenes. The framework is first described at a conceptual level where we show that it can, in principle, account for a range of psychophysical, anatomical, and physiological phenomena. Then we describe some specific models (parts of the framework which we have developed in detail) and show that they agree in detail with some of these phenomena. The framework suggests a number of experimental and computational questions, which must be answered as a complete theory There are many theories of motion measurement in both the biological (for reviews, see Smith and Snowden 1994) and artificial-vision (Blake and Yuille 1992, Faugeras 1993) literatures. Typically, each theory is only suitable for describing limited types of motion. However, the growing success of such theories leads us to ask the big