Evaluating quality of surface description using robust methods

In previous work (Koivunen and Pietikainen, 1991) the authors presented a segmentation method that combines useful properties of edge and region-based segmentation. The least squares estimation used gives good results when pixels in the neighborhood are from one statistical population, and the noise is Gaussian distributed. To be able to deal with very deviant pixel values, the authors applied an iterative reweighting least squares method and a least trimmed squares method for surface description. This paper presents the improvements on the robustness of the surface description, and a quantitative analysis of the quality of the description. The validity of the assumptions used is also evaluated quantitatively. Both synthetic and real range images are used for test images