Fuzzy Random Walk

Abstract

Random walks applied to radiosity use random lines to simulate the distribution of luminous power. Noise effects in the final image are inherent in these methods. We present an improvement that reduces this noise, avoiding the sharp transitions between every patch and its coplanar neighbours, and obtaining a smoother final image with a negligible increase in computational cost.

The main idea involved in our improvement is the distribution of a part of the incoming to a patch power between its neighbours. That is, part of the power splats the neighbour patches. Our approach is based on Information Theory concepts, as uncertainty. Reductions of the mean square error to less than a half have been obtained in our tests.