Title:Inference based on Procrustes means for the 3D shape of tree boles and mildly rank-deficient diffusion tensors

Abstract: A method for inference on cylindricity of tree boles based on 3D Kendall shapes is presented. This method employs Procrustes means which are highly popular in the shape community. While for extrinsic means (such are Procrustes means for 2D Kendall shapes) and intrinsic means, a strong law of large numbers (SLLN) and a central limit theorem (CLT) are available, Procrustes means, however as goes the common belief due to a perturbation paradigm, could not be used for inference because they are "consistent estimators" only under specific distributional assumptions. In this work, the impact of shape space geometry on this paradigm is discussed in detail. To this end, SLLN and CLT are extended to a generalization of Fréchet means, in particular making three- and higher-dimensional Procrustes means and a computationally slightly less costly mean introduced by Ziezold (1994) available for asymptotic non-parametric inference. As another application, a framework for inference on mildly rank-deficient diffusion tensors as occur in neuro-imaging is established.