An Exploration of Stochastic Completion Fields

We present in this paper a method originally proposed by Williams and Jacobs to determine the shape of illusory contours. The theory on stochastic completion fields is based on the assumption that the prior probability distribution of completion shapes can be modelled by the random walks of particles. Two different algorithms to compute completion fields are presented. The first is based on the convolution of a large kernel determined by MonteCarlo simulation and the other on the integration of the Fokker-Planck equation by repeated convolutions of small kernels. Experimental results are then presented to demonstrate the effectiveness of stochastic completion fields. We conclude with a discussion that suggests when the use of completion fields may fail.