Minkowski tensors for point clouds and voxelized data: robust, asymptotically unbiased estimators

D. Hug, M.A. Klatt, D. Pabst · 2025

Read Minkowski tensors for point clouds and voxelized data: robust, asymptotically unbiased estimators by D. Hug, M.A. Klatt, D. Pabst (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-00092.txt.

Introduction

Daniel Hug, Michael A Klatt, and Dominik Pabst improve tools that measure the shape of random spatial structures from point clouds and voxel data. Their 2025 paper works with Minkowski tensors, summaries that capture multi-point shape information and resist noise. Simple local estimators of these tensors stay biased even as resolution grows without limit.

The authors sharpen a recent bias-free algorithm, making it sturdier and faster, and widen its theory to cover shapes like rough surfaces and composites. They test it on random polytopes, where exact expected values can be derived, and land within a few percent at practical resolutions. Applications to metal grains and nano-rough surfaces follow, with an open-source package released for any dimension.

Catalog record

CREATORD. Hug, M.A. Klatt, D. Pabst
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE128 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-00092

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Minkowski tensors for point clouds and voxelized data: robust, asymptotically unbiased estimators D. Hug, M.A. Klatt, D. Pabst Abstract Minkowski tensors, also known as tensor valuations, provide robust n n -point information for a wide range of random spatial structures. Local estimators for point clouds, e.g., representing voxelized data, however, are unavoidably biased even in the limit of infinitely high resolution. Here, we substantially improve a recently proposed, asymptotically unbiased algorithm to estimate Minkowski tensors from point clouds. Our improved algorithm is more robust and efficient. Moreover we generalize the theoretical foundations for an asymptotically bias-free estimation of the interfacial tensors, among others, to the case of finite unions of compact sets with positive reach, which is relevant for many applications like rough surfaces or composite materials. As a realistic test case of random spatial structures, we consider random (beta) polytopes. We first derive explicit expressions of the expected Minkowski tensors, which we then compare to our simulation results. We obtain precise estimates with relative errors of a few percent for practically relevant resolutions. Finally, we apply our methods to real data of metallic grains and nanorough surfaces, and we provide an open-source python package, which works in any dimension. Keywords: tensor valuations, microstructure characterization, anisotropy, digitized image analysis 2020 MSC: 94A08 · 68U10 · 60D05 · 53C65 · 28A75 · 62H35 · 52A22 1 Introduction Random spatial structures appear ubiquitously in nature and technology. Examples (and their corresponding mathematical models) are heterogeneous materials and porous media (random sets) [ 57 , 78 , 58 , 1 , 6 ] , cellular tissues and foam-like structures (tessellations) [ 59 , 64 , 42 , 74 ] , rough surfaces (random fields) [ 2 , 79 , 73 , 65 ] , and bacterial colonies (particle processes) [ 17 , 26 , 81 ] as well as error-correcting code (high-dimensional sphere packings) [ 82 , 19 ] . This variety of examples corresponds to a similar diversity in random shapes. In all of these cases, Minkowski tensors (also known as tensor valuations) [ 67 , 36 ] provide a comprehensive, unified shape analysis. The Minkowski tensors [ 70 ] are generalizations of the well-known Minkowski functionals (or intrinsic volumes) from integral geometry [ 67 ] ; see Section 2 for more details and the theoretical background. The intrinsic volumes represent intuitive geometric information, e.g., on the volume, surface area, or curvature; the Minkowski tensors additionally distinguish different orientations with respect to these geometrical properties [ 44 ] ; see also Fig. 1 . Figure 1: Distinct Minkowski tensors quantify different types of anisotropy: (a–b) moment tensors of the volume and boundary distributions capture the anisotropy of disk arrangements; (c) the Euler characteristic \chi is a topological constant, generalizations to surface tensors \Phi_{0}^{0,2s} are proportional to \chi and the metric tensor and hence always isotropic, see ( 2.8 ); (d–f) anisotropy of the surface‑normal distribution is measured by surface tensors, different symmetries are captured by tensors of a corresponding rank; (g) curvature anisotropy is incorporated by an additional curvature function to the surface tensors. Minkowski tensors of different ranks capture symmetries of all orders. They quantify the degree of anisotropy as well as the preferred directions. They are comprehensive in the sense of Hadwiger’s and Alesker’s characterization theorems [ 67 ] , intuitively speaking, they contain all motion-covariant, continuous, and additive shape information. Since they are additive, Minkowski functionals and tensors provide robust access to n -point information [ 53 ] . Moreover, the definition in real space allows for a convenient treatment of boundary conditions [ 72 ] . The versatility of Minkowski functionals and tensors is demonstrated by their ample successful applications in a wide range of fields, including statistical physics [ 54 , 55 , 45 ] , biology [ 11 , 10 , 9 ] , spatial statistics [ 81 , 23 ] , image analysis [ 57 , 58 ] , as well as astronomy and cosmology [ 25 , 66 , 38 , 43 , 21 , 20 ] . Explicit examples of random spatial structures characterized via Minkowski functionals and tensors are porous media [ 7 , 8 , 41 , 6 , 37 ] , composite materials [ 24 ] , nanorough surfaces [ 73 ] , fluid demixing [ 13 ] , and trabecular bone [ 64 , 44 , 15 ] . A challenge, however, is the application of Minkowski tensors to point clouds, e.g., representing voxelized data, where a systematic bias can often not be avoided even in the limit of infinitely high resolution. In contrast, for voxelized gray-scale data, marching-cube algorithms can create triangulated meshes with continuous orientations, which results in asymptotically bias-free algorithms [ 51 , 50 , 75 , 76 ] . If no gray-scale data is available, any local algorithm for Minkowski tensors using black-and-white voxels will be asymptotically biased, i.e., the estimator will not converge to the true value in the limit of infinite resolution [ 77 ] . An algorithmic solution to the problem was recently proposed in [ 32 ] , which had been inspired by [ 16 , 56 ] . A more general framework, admitting to work with general distance-like functions (such as power distances and, correspondingly, power cells) to improve the robustness of estimation procedures against outliers, while preserving the resilience with respect to Hausdorff noise, was explored in [ 22 ] ; for a discussion of and references to related methods, see [ 22 ] . The key idea of the algorithmic solution in [ 32 ] is to construct Voronoi cells that can encode global information and thus guarantee asymptotically bias-free estimators of all Minkowski tensors for sets of positive reach. This requirement of a positive reach, however, excludes many interesting applications, like rough surfaces and composite materials. More importantly, the proposed algorithm suffered from numerical instabilities since it relies on a matrix inversion, which is slow and sensitive to rounding errors, especially for ill‑conditioned matrices. Here, we substantially advance the robustness of the algorithm by replacing the matrix inversion by a least squares fit. The latter avoids numerical instabilities and allows for more data from parallel sets than needed, which further enhances robustness against statistical and systematic errors. Furthermore, we increase its efficiency by using an unbiased estimator of the Voronoi tensors (also called Voronoi volume integrals). This revised algorithm, here called Voronoi-LSQ algorithm, is also asymptotically unbiased for sets of positive reach, as we rigorously prove; see Remark 4.13 . We demonstrate its robustness and accuracy for exemplary cases of geometric shapes, stochastic models, and real data. Even for a set without positive reach, our algorithm achieves surprisingly good results with relative deviations of less than 3% in the eigenvalues of the interfacial tensors, see Table 6 , at least for the simple example of a rectangular shell. Since such precision is not theoretically substantiated, we also introduce an alternative method, here called Voronoi-FD algorithm, to estimate interfacial tensors, for which we are able to prove that its corresponding estimator is asymptotically unbiased for a large class of sets, including parallel sets of compact sets or finite unions of compact sets with positive reach (see Theorem 4.8

Read the complete text