Réduction de Graphes et Application la Segmentation de Tumeurs Pulmonaires

Nicolas Lerme

Description

Niveau: Supérieur, Doctorat, Bac+8
Réduction de Graphes et Application à la Segmentation de Tumeurs Pulmonaires Réduction de Graphes et Application à la Segmentation de Tumeurs Pulmonaires Soutenance de thèse Nicolas Lermé LAGA UMR CNRS 7539 / LIPN UMR CNRS 7030 Université Paris 13 December 7, 2011 Nicolas Lermé Soutenance de thèse 1 / 54

• segmentation de tumeurs pulmonaires


• photo edition


• reducing graphs


• lipn umr


• image segmentation


• ill-posed inverse

Sujets

• Segmentation (image processing)

Informations

• Publié par: mijec
• Nombre de lectures: 45
• Langue: Français
• Poids de l'ouvrage: 6 Mo

Extrait

RéductiondeGrapehesAtppilacitnoaSàlmeegatntndiomuTesruemluPianores tuoSémreLsalociN/5e1èsthdeceanen

Réduction de Graphes et Application à la Segmentation de Tumeurs Pulmonaires Soutenance de thèse

LAGA UMR CNRS 7539 / LIPN UMR CNRS 7030 Université Paris 13

4

Nicolas Lermé

December 7, 2011

1

Preliminaries

tenaéSouethènced45

Outline

es/2

GCSFMP: an application of RGC for segmenting lung tumors

4

Conclusion and future work

5

2

Reducing graphs: state-of-the-art

Reduced Graph Cuts

3

arhpsetepAlpcitaRéductiondeGemuTednoomluPsruSelanàiotitaengmesnair LermolasNic

tAselippticaàlongeSatnemoitaTednRdécuitnoedrGpaehiaermlnosruPmuue esnarilimi s Pre eLmrlosaetanSéuoNic

Reduced Graph Cuts

3

2

Reducing graphs: state-of-the-art

Conclusion and future work

5

GCSFMP: an application of RGC for segmenting lung tumors

4

Preliminaries

1

ednchèet3/se

Outline

54

salociNuoSémreLedncnate4/sehèet

Objective

54

(-) Well known ill-posed inverse problem. (+) Intensively studied since 70’s. Applications: medical diagnosis, photo edition, etc.

Figure 1:Examples of segmentations.

Partition an image in disjoints homogeneous regions according to some criteria.

Remarks

rsPuumeundeTatioemtnSagenolàacitlipptAseheapGrdenoitcudéR emionproblgeemtnta no mIgase odtrtiuc rina In eserP s imil nomleria

PsruemuTednoitatengmSelanàioatic udtctnor ei I sniraelim es Pr nairulmosetepAlpdnGearhpRéductio ioatppnaacrohE noi grennimyzimi ociNsLlaméerutSoanen

(1)

E p,q (0,0) +E p,q (1,1)≤E p,q (0,1) +E p,q (1,0).

Remarks Minimizing (1) is NP-hard. (1) can be solved in polynomial-time whenE p,q (.)is submodular:

amongu∈ {0,1} P and where Nis a standard neighborhood system. E p (.)andE p,q (.)depend on inputs (image and user interaction). βis a regularization parameter.

We want to minimize the following energy: E(u) =β X E p (u p ) + X E p,q (u p ,u q ), p∈P (p,q)∈N

4/5e5èsthdece