Segmentation of Range Image by Using Restricted Quadric Surfaces

Abstract

In this paper, a clustering method is proposed to perform segmentation for range image of the object that has arbitrary topology. This approach consists of two stages in order to determine desirable clusters. First, while new clusters are created on demand, the parameters of each created cluster are estimated in recursive form (Stage 1).As the algorithm processes, next, the clusters that have been created at Stage 1 compete each other, and clusters that lose the competition gradually vanish (Stage 2). Consequently, since the only desirable clusters are left, the segmentation can be performed by assigning the data to the left clusters.