kgk@CS.BROWN.EDU (11/23/88)
Parallel Computation of Motion: computation, psychophysics and physiology Heinrich H. B"ulthoff Brown University Department of Cognitive and Linguistic Sciences Wednesday, November 30, 12PM. Lubrano Conference Room 4th Floor, Center for Information Technology Brown University The measurement of the 2D field of velocities -- which is the projection of the true 3D velocity field -- from time-varying 2-dimensional images is in general impossible. It is, however, possible to compute suitable ``optical flows'' that are qualitatively similar to the velocity field in most cases. We describe a simple, parallel algorithm that computes successfully an optical flow from sequences of real images, is consistent with human psychophysics and suggests plausible physiological models. In particular, our algorithm runs on a Connection Machine supercomputer in close to real time. It shows several of the same ``illusions'' that humans perceive. A natural physiological implementation of the model is consistent with data from cortical areas V1 and MT. Regularizing optical flow computation leads to a formulation which minimizes matching error and, at the same time, maximizes smoothness of the optical flow. We develop an approximation to the full regularization computation in which corresponding points are found by comparing local patches of the images. Selection between competing matches is performed using a winner-take-all scheme. The algorithm accomodates many different image transformations uniformly, with similar results, from brightness to edges. The algorithm is easily implemented using local operations on a fine-grained computer (Connection Machine) and experiments with natural images show that the scheme is effective and robust against noise. This work was done at the Artificial Intelligence Laboratory at MIT in collaboration with Tomaso Poggio and James J. Little .