Hierarchical segmentation of range images with contour constraints

This paper describes a new algorithm to segment in continuous parametric regions range images. The algorithm starts with an initial partition of small first order regions using a robust fitting algorithm constrained by the detection of depth and orientation discontinuities. The algorithm then optimally group these regions into larger and larger regions using parametric functions until an approximation limit is reached. The algorithm uses Bayesian decision theory to determine the local optimal grouping and the complexity of the parametric model used to represent the range signal. After the segmentation process an exact description of the boundary of each region is computed from the mutual intersections of the extracted surfaces. Experimental results show significant improvement of region boundary localization. A systematic comparison of our algorithm to the most well known algorithm in the literature is presented to highlight the contributions of this paper.