A neural network model of long-range apparent motion in the human vision

Detection of motion is one of the most important functions of the visual systems of animals. Motion perceived by humans when succeeding frames are presented is called 'apparent motion'. According to its psychophysical characteristics, a categorization of apparent motion into two parts has been proposed, i.e. the short-range apparent motion (SRAM) and the long-range one (LRAM). From a computational viewpoint, SRAM and LRAM are characterized by filtering and matching tasks, respectively. In this paper, a neural network model of LRAM is proposed. The model uses a neural network of the Hopfield type which solves the matching task as an optimization problem. Computer simulation shows that the network yields a satisfactory solution of the matching task with careful consideration of the energy function. Also, it is shown that the number of iterations required in the computation is in a plausible range for a model of human visual processing.<>