On 3-D image segmentation based on planar and quadric approximations

In this paper we propose an algorithm for segmentation of range data by planar and quadric region extraction using the region based approach. The algorithm is a combination of split-and-merge and region growing algorithms. Planar regions are extracted first by clustering approach, wherein, during each iteration the co-ordinate along which the separation of regions is highest is selected. This simplifies the multidimensional clustering problem to a one-dimensional clustering problem, which is computationally efficient. Region growing approach has been implemented for non-planar regions using the quadric fitting