Towards Morse Theory for Point Cloud Data

نویسندگان

  • F. Cazals
  • A. Roth
چکیده

Morse theory provides a powerful framework to study the topology of a manifold from a function de ned on it, but discrete constructions have remained elusive due to the di culty of translating smooth concepts to the discrete setting. Consider the problem of approximating the Morse-Smale (MS) complex of a Morse function from a point cloud and an associated nearest neighbor graph (NNG). While following the constructive proof of the Morse homology theorem, we present novel concepts for critical points of any index, and the associated Morse-Smale diagram. Our framework has three key advantages. First, it requires elementary data structures and operations, and is thus suitable for high-dimensional data processing. Second, it is gradient free, which makes it suitable to investigate functions whose gradient is unknown or expensive to compute. Third, in case of under-sampling and even if the exact (unknown) MS diagram is not found, the output conveys information in terms of ambiguous ow, and the Morse theoretical version of topological persistence, which consists in canceling critical points by ow reversal, applies. On the experimental side, we present a comprehensive analysis of a large panel of bi-variate and tri-variate Morse functions whose Morse-Smale diagrams are known perfectly, and show that these diagrams are recovered perfectly. In a broader perspective, we see our framework as a rst step to study complex dynamical systems from mere samplings consisting of point clouds. Key-words: Morse theory, Morse homology, point cloud data, multi-scale analysis, energy landscapes ∗ Inria Sophia-Antipolis † ETH Zurich ‡ CNRS § Inria Sophia-Antipolis ha l-0 08 48 75 3, v er si on 1 28 J ul 2 01 3 Vers une théorie de Morse pour les nuages de points Résumé : La théorie de Morse fournit un formalisme puissant pour étudier une variété à partie d'une fonction de nie sur celle-ci, mais généraliser au cas discret les constructions connues en topologie di érentielle est une problématique ouverte. Considérons le problème consistant à approximer le diagramme de Morse-Smale (MS) d'une fonction à partir d'une graphe de plus proches voisins (NNG) dé ni sur un nuage de point échantillonant cette fonction. En suivant la construction du theoreme central de l'homologie de Morse, nous présentons des concepts nouveaux de points critiques, ainsi que le diagramme de Morse-Smale associé. Notre canevas présente trois avantages clefs. Tout d'abord, les primitives utilisées relèvent d'algorithmes et structures de données élémentaires, adaptées au traitement de données en grande dimension. Ensuite, la connaissance du gradient de la fonction étudiée n'est pas requis. En n, en cas de sous-echantilonnage et même si le diagramme de MS exact n'est pas retrouvé, des informations sur des ambiguités de ot sont mises en évidence. Du point de vue expérimental, nous présentons une analyse exhaustive de fonctions 2D et 3D dont les diagrammes de MS sont connus, et parfaitement retrouvés. Dans un registre plus général, nous pensons que notre canevas est un premier pas dans la perspective de l'étude de systèmes dynamiques à partir de nuages de points. Mots-clés : Théorie de Morse, Homologie de Morse, nuages de points, analyse multi-echelle, surfaces d'énergie ha l-0 08 48 75 3, v er si on 1 28 J ul 2 01 3 Towards Morse Theory for Point Cloud Data 3

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Baby Morse Theory for Statistical Inference from Point Cloud Data

A methodology is proposed for inferring the topology underlying point cloud data directly from point cloud data. The approach employs basic elements of Morse theory, and is capable of producing not only a point estimate of various topological quantities (e.g., genus), but it can also assesses their sampling uncertainty in a probabilistic fashion. Several examples of point cloud data in three di...

متن کامل

Computational Topology in Text Mining

In this paper we present our ongoing research on applying computational topology to analysis of structure of similarities within a collection of text documents. Our work is on the fringe between text mining and computational topology, and we describe techniques from each of these disciplines. We transform text documents to the so-called vector space model, which is often used in text mining. Th...

متن کامل

Conditional Random Fields for Airborne Lidar Point Cloud Classification in Urban Area

Over the past decades, urban growth has been known as a worldwide phenomenon that includes widening process and expanding pattern. While the cities are changing rapidly, their quantitative analysis as well as decision making in urban planning can benefit from two-dimensional (2D) and three-dimensional (3D) digital models. The recent developments in imaging and non-imaging sensor technologies, s...

متن کامل

Computing Persistent Homology via Discrete Morse Theory

This report provides theoretical justification for the use of discrete Morse theory for the computation of homology and persistent homology, an overview of the state of the art for the computation of discrete Morse matchings and motivation for an interest in these computations, particularly from the point of view of topological data analysis. Additionally, a new simulated annealing based method...

متن کامل

A novel method for locating the local terrestrial laser scans in a global aerial point cloud

In addition to the heterogeneity of aerial and terrestrial views, the small scale terrestrial point clouds are hardly comparable with large scale and overhead aerial point clouds. A hierarchical method is proposed for automatic locating of terrestrial scans in aerial point cloud. The proposed method begins with detecting the candidate positions for the deployment of the terrestrial laser scanne...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2013