Skip to Main content Skip to Navigation

Squelettisation d’images en niveaux de gris et applications

Abstract : Skeletonization is an image transformation that aims to represent objects by their medial axis while preserving their topological characteristics (homotopy). It is widely used in biometrics, character recognition and also in the extraction of bone microarchitecture. The objective of this thesis is to develop a skeletonization method applied directly on image gray levels. This has the large advantage of freeing the operation from preprocessing techniques such as binarization. A review of grayscale skeletonization methods shows that the morphological thinning is one of the most used approaches for its topology preservation property. However, this approach is sensitive to image noise and produces inexploitable skeletons. A first parameterization of the thinning process has been proposed in the literature to reduce noise-related information. The first contribution of this work is to propose an adjustment of this parameter based on a statistical decision. To this end, a hypothesis test is identified for each lowering criterion in order to set the thinning parameter locally. This leads us to propose the Self Contrast Controlled Thinning method SCCT that is robust to noise and is automatically adjusted to image contrast. The SCCT is made available to application domains through its optimized implementation based on hierarchical queues. Noticing the lack of efforts to assess grayscale skeletonization, the second contribution of this work is to propose a quantitative evaluation protocol assessing skeletonization with regard to its fundamental properties that are namely the preservation of topology and geometry. This protocol is conducted over a synthetic images database and allows us to compare SCCT to approaches from the literature. The third contribution consists in structuring the skeleton into a graph that gives access to objects structural and morphometric descriptors and enables the exploitation of the skeleton by experts from various fields of application. This structuring is applied in the context of Voxelo project which is coordinated by B2OA laboratory of the University Paris Diderot. In this context, descriptors of bone microarchitecture quality are extracted from X-ray high resolution images.
Document type :
Complete list of metadata

Cited literature [130 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Monday, August 28, 2017 - 4:18:24 PM
Last modification on : Saturday, June 19, 2021 - 3:10:51 AM


Version validated by the jury (STAR)


  • HAL Id : tel-01578084, version 1


Rabaa Douss. Squelettisation d’images en niveaux de gris et applications. Logiciel mathématique [cs.MS]. Université Sorbonne Paris Cité; Université de Carthage (Tunisie), 2015. Français. ⟨NNT : 2015USPCB138⟩. ⟨tel-01578084⟩



Les métriques sont temporairement indisponibles