Surface approximation and range image segmentation through robust competitive clustering

Algorithms that perform segmentation and surface approximation of range images need to be robust since real range data tends to be noisy. In this paper, we present a robust clustering algorithm and show how it can be used to obtain an approximation of a range image in terms of quadric surface patches. The proposed algorithm does not assume that the number of surface patches is known a priori, and performs well even when the data set is contaminated by noise and outliers.