A framework for constrained multiscale range image segmentation

In this paper, we present a general framework for multiscale segmentation of range images in planar facets. We propose a formulation of the problem as the minimization of a multiscale separable energy, which involves two terms: a goodness-of-fit term and a regularization term, the scale parameter being the relative weight between the two terms. We then recall a general parameter-free method to build a hierarchy of regions in which segmentation is expressed as a cut. Afterwards, we explicit the energy derived for a simple-shaped planar facet model and based on a formulation that confronts the complexity of the model and the goodness of fit of the data. We then extend this model to integrate gradients information from images, which leads us to a formulation of a simple-shaped planar and homogeneous facet model. Some results on digital elevation models (DEM) obtained from aerial images with and without image constraints are provided.