Computing a discrete Morse gradient from a watershed decomposition. (August 2016)

• Record Type: Journal Article
• Authors: Čomić, Lidija; De Floriani, Leila; Iuricich, Federico; Magillo, Paola
• Is Part Of: Computers & graphics. Volume 58(2016)
• Page Start: 43
• Page End: 52
• Publication Date: 2016-08
• Subjects: Discrete Morse theory -- Watershed transform -- Morse–Smale complexes; Computer graphics -- Periodicals; 006.6
• DOI: 10.1016/j.cag.2016.05.020
• Languages: English
• ISSNs: 0097-8493

Abstract

Abstract: We consider the problem of segmenting triangle meshes endowed with a discrete scalar function f based on the critical points of f . The watershed transform induces a decomposition of the domain of function f into regions of influence of its minima, called catchment basins. The discrete Morse gradient induced by f allows recovering not only catchment basins but also a complete topological characterization of the function and of the shape on which it is defined through a Morse decomposition. Unfortunately, discrete Morse theory and related algorithms assume that the input scalar function has no flat areas, whereas such areas are common in real data and are easily handled by watershed algorithms. We propose here a new approach for building a discrete Morse gradient on a triangulated 3D shape endowed by a scalar function starting from the decomposition of the shape induced by the watershed transform. This allows for treating flat areas without adding noise to the data. Experimental results show that our approach has significant advantages over existing ones, which eliminate noise through perturbation: it is faster and always precise in extracting the correct number of critical elements. Abstract : Graphical abstract: Abstract : Highlights: An algorithm for computing a Forman gradient on scalar fields with flat areas. Equivalence of the techniques based on discrete Morse theory and watershed. Critical simplexes computed in 1-to-1 correspondence with the critical points. Source code of the proposed method publicly available.