Fuzzy surface descriptions for 3-D machine vision

Traditional curvature measures for modeling 3-D surface classify surfaces into crisp sets based on the sign of the mean and Gaussian curvatures. However, descriptions based on such measures do not represent the intuitive descriptions in a natural way, i.e. the degree to which the segment belongs to each of the surface types in the crisp set. In addition, curvature estimates are extremely sensitive to noise due to the computation of directional derivatives, which makes classification more difficult. There exists a certain level of uncertainty/ambiguity that is not taken into account while classifying the surfaces based on the existing methods. In this paper, a novel fuzzy surface description technique, that emulates the natural description of surfaces, is proposed and demonstrated on a class of range images