Abstract

Minimal partition problems consist in finding a partition of a domain into a given number of components in order to minimize a geometric criterion. In applicative fields such as image processing or continuum mechanics, it is standard to incorporate in this objective an interface energy that accounts for the lengths of the interfaces between components. The present work is focused on thetheoretical and numerical treatment of minimal partition problems with interface energies. The considered approach is based on a Gamma-convergence approximation and duality techniques.

Highlights

  • 1.1 Problem descriptionConsider a partition of a bounded domain D of R2 into relatively closed subsets Ω1, . . . , ΩN called phases that may intersect only through their boundaries:D = ∪Nj=1Ωj, and Ωi ∩ Ωj = ∂Ωi ∩ ∂Ωj ∩ D for i = j.Denote the interface separating Ωi and Ωj by Γij: Γij = ∂Ωi ∩ ∂Ωj ∩ D for i = j, with the additional convention Γi,i = ∅

  • In particular we look at binary and multilabel minimal partition problems, including supervised or automatic image classification and deblurring

  • We show that the interface energy can be rewritten as a linear combination of perimeters and we are able to compute the coefficients

Read more

Summary

21 Septembre 2017

Je tiens à remercier Zakaria BELHACHMI et Édouard OUDET d’avoir accepté de rapporter sur ma thèse. Je voudrais aussi remercier Luc DINH THE et Jérôme FEHRENBACH d’avoir accepté de faire partie de mon jury de thèse. Un grand merci à Samuel AMSTUTZ pour ses conseils, recommandations et encouragements. Merci pour être toujours disponible et pour avoir une réponse à toutes mes questions. Je voudrais remercier tous les membres du LMA. Je remercie Anna et Chiara pour les bons moments qu’on a passés ensemble. Je veux remercier ma famille pour m’avoir encouragé à suivre ma passion pour les mathématiques. Merci Atika, Israa, Zahraa et Athraa pour être toujours près de moi. Tout ce que je fais c’est pour vous

Problem description
Motivation where αij is a coefficient called surface tension associated with
Image processing
Material sciences
Relaxation
Convexification
Γ-convergence
The present work
Thesis outline
Thesis outline Chapter 5
Some results from integration
Weak differentiation
The space H1/2(∂D)
The space Hdiv(D)
Linear operators Throughout this subsection, E and F denote two Banach spaces
Basic geometric measure theory
BV space
Sets of finite perimeter
Γ−convergence
Legendre-Fenchel transform (ii) If G : X → R is continuous, then (Fn + G)n∈N Γ−converges to F + G
Legendre-Fenchel transform
The Lax-Milgram theorem
Homogeneous Neumann problem
The operator Lε
The Modica Mortola functional
A gradient-free perimeter approximation
Mathematical results
Lower semicontinuity iteration of an outer loop, that is, we choose εi = εmax/2i
Numerical examples
Lower semicontinuity
Equicoercivity
Equicoercivity (a)
Algebraic properties of interface energies
Reformulation in terms of perimeters and thus
Matrix representation of algebraic properties
Existence of conical combination
Γ−convergence with non negative coefficients
Primal algorithm
Extension: volume constraints
Primal dual formulation
Primal dual algorithm
Saddle point formulation Let us start by defining the functional
Saddle point algorithm
Comparison between saddle point and primal algorithms
Numerical results
Condition for concavity: conditional negative definiteness
Concavity of the approximate interface energy
Dual formulation of the interface energy by Legendre-Fenchel transform
Dual formulation in the discrete framework
Algorithm
Image classification (c) Primal algorithm for α23=1/2 and αij =1 otherwise
Classification of color images
Projected gradient algorithm
Projected gradient algorithm ξ3
Image deblurring of greyscale image
Image deblurring of colour image (a) Damaged image with blur and noise effects
Image deblurring of colour image
Medical imaging
Full Text
Paper version not known

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call

Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.