QUALITATIVE PRIMITIVE IDENTIFICATION USING FUZZY CLUSTERING AND INVARIANT APPROACH

This paper addresses the issue of qualitative primitive extraction fro m range images. The fuzzy C-shell clustering technique is applied to p artition range images into a set of quadric shells. Once the best-fitt ing quadric shells have been recovered, five types of invariants are u sed to classify the qualitative shapes and represent the geometric par ameters of the shapes. Finally, the extracted quadric shells are local ised in terms of quadric centre(s), quadric directions, and any axes o f revolution. Using fuzzy shell clustering, the shell features can be segmented and fitted simultaneously, and individual best-fitted shells can be clustered concurrently. The integration of the partition with the invariant analysis makes it possible to identify qualitative featu res from depth maps.