Project

Back to overview

Delineating Trees in Noisy 2D Images and 3D Image-Stacks

English title Delineating Trees in Noisy 2D Images and 3D Image-Stacks
Applicant Fua Pascal
Number 149866
Funding scheme Project funding (Div. I-III)
Research institution Laboratoire de vision par ordinateur EPFL - IC - ISIM - CVLAB
Institution of higher education EPF Lausanne - EPFL
Main discipline Information Technology
Start/End 01.10.2013 - 31.07.2016
Approved amount 219'850.00
Show all

Keywords (3)

Delineation; Computer Vision; Dendritic Trees

Lay Summary (French)

Lead
Les structures arborescentes apparaissent à différentes échelles et dans différents contextes. A l’échelle micrométrique ce peut être des dendrites observées au microscope optique, à l’échelle centimétrique des vaisseaux sanguins dans des fonds de rétines, et à l’échelle métrique des routes dans des photos aériennes. Les extraire de manière automatique et robuste est donc d’une importance fondamentale pour de nombreuses disciplines scientifiques.
Lay summary

Cependant, bien que le sujet ait reçu une attention soutenue depuis les débuts de la  vision par ordinateur, la robustesse et l’automatisation restent élusives. Il existe des techniques entièrement automatiques mais elles ne produisent des données satisfaisantes que pour des données non bruitées. Toutes les autres requièrent beaucoup d’intervention manuelle.

Dans ce projet, nous allons donc mettre en place une approche entièrement automatisée pour combler ces lacunes. Nous allons d’abord proposer une fonction objectif pour évaluer les arbres en termes à la fois des données images et les propriétés géométriques connues des structures que nous cherchons. Nous mettrons ensuite en œuvre des algorithmes pratiques pour construire des arbres qui maximise cette fonction objectif en un temps raisonnable même si le problème sous-jacent est étroitement lié à un problème connu pour être NP-difficile.

Direct link to Lay Summary Last update: 30.09.2013

Responsible applicant and co-applicants

Employees

Publications

Publication
Active Learning for Delineation of Curvilinear Structures
Mosinska A., Sznitman R., Glowacki P., Fua P. (2016), Active Learning for Delineation of Curvilinear Structures, in Conference on Computer Vision and Pattern Recognition, Las Vegas, NEIEEE, Las Vegas, NE.
Reconstructing Evolving Tree Structures in Time Lapse Sequences
Glowacki P., Pinheiro M.A., Turetken E., Sznitman R., Lebrecht D., Holtmaat A., Kybic J., Fua P. (2014), Reconstructing Evolving Tree Structures in Time Lapse Sequences, in Conference on Computer Vision and Pattern Recognition, Columbus, OHIEEE, Columbus, OH.
Reconstructing Loopy Curvilinear Structures Using Integer Programming
Turetken E., Benmansour F, Andres B., Pfister H., Fua P. (2013), Reconstructing Loopy Curvilinear Structures Using Integer Programming, in Conference on Computer Vision and Pattern Recognition, Portland, ORIEEE, Portland, OR.
Reconstructing Curvilinear Networks Using Path Classifiers and Integer Programming
Turetken E., Benmansour F., Andres B., Glowacki P., Pfister H., Fua P., Reconstructing Curvilinear Networks Using Path Classifiers and Integer Programming, in IEEE Transactions on Pattern Analysis and Machine Intelligence, 1.

Collaboration

Group / person Country
Types of collaboration
University of Geneva Switzerland (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Exchange of personnel
KU Leuven Belgium (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Exchange of personnel

Associated projects

Number Title Start Funding scheme
134734 Delineating Trees in Noisy 2D Images and 3D Image-Stacks 01.09.2011 Project funding (Div. I-III)
131549 Advanced Learning for Tracking and Detection in Medical Workflow Analysis. 01.04.2012 Project funding (Div. I-III)
127456 Understanding Brain morphogenesis: Computer Vision Morphological Feature Extraction and 01.10.2009 Sinergia
121737 Automating Delineation of Dendritic Networks in Noisy Image Stacks 01.09.2009 Project funding (Div. I-III)
130470 Neural Circuit Reconstruction 01.01.2011 Sinergia
166107 Domain Adaptation for Microscopy Imaging 01.08.2016 Project funding (Div. I-III)
166107 Domain Adaptation for Microscopy Imaging 01.08.2016 Project funding (Div. I-III)

Abstract

Tree-like structures appear at many different scales and in many different contexts. They can be micrometer scale dendrites in light microscopy image-stacks, centimeter-scale blood vessels in retinal scans, or meter-scale road networks in aerial images. Extracting them automatically and robustly is therefore of fundamental relevance to many scientific disciplines. However, even though the topic has received sustained attention ever since the inception of the field of Computer Vision, both robustness and automation remain elusive. Fully automated techniques exist but require very clean data; substantial amounts of manual intervention is required for any other kind.In this project, we will therefore develop a fully automated approach to addressing these shortcomings. We will first develop an approach to finding optimal trees that accounts both for global image and geometric properties. We will then implement a practical algorithm to build near-optimal ones in an acceptably short time, even though the underlying problem is closely related to one known to be NP-Hard.
-