Ground Truth Correction for Topologically Better Delineation of Curvilinear Structures
Event details
| Date | 18.06.2019 |
| Hour | 09:00 › 11:00 |
| Speaker | Doruk Oner |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Bertrand Merminod
Thesis advisor: Prof. Pascal Fua
Co-examiner: Dr. Graham Knott
Abstract
Delineation of curvilinear structures is a popular direction of research in computer vision since such structures are used in many real world problems like biomedical image analysis. The idea is to find centerlines of the desired structures in a given image while preserving the overall topology of the structure. In this proposal, we analyzed three works that are related to this problem. We presented the problems they solved, the methods they used, results of the works and how we can integrate these work to our research. The first paper focuses on an end-to-end 3D convolutional network to produce individual segments. The second paper is a survey paper on deformable models, their extensions and applications. The third paper is on a new method on unsupervised reconstruction of 3D point clouds. Finally, we talked about our current research, and future plans and discussed how to use these papers to improve our research to obtain topologically better capture of curvilinear structures.
Background papers
Flood-fliing Networks by Michal Januszewski, by Jeremy Maitin-Shepard, Peter Li, Jörgen Kornfeld, Winfried Denk and Viren Jain, CoRR 2016.
Deformable models in medical image analysis: a survey by Tim McInerney and Demetri Terzopoulus, Medical Image Analysis (MIA) 1996.
FoldingNet: Point Cloud Auto-Encoder via Deep Grid Deformation, by Yaoqing Yang, Chen Feng, Yiru Shen and Dong Tian, IEEE/CVF Conference on Computer Vision and Pattern Recognition 2017.
Exam president: Prof. Bertrand Merminod
Thesis advisor: Prof. Pascal Fua
Co-examiner: Dr. Graham Knott
Abstract
Delineation of curvilinear structures is a popular direction of research in computer vision since such structures are used in many real world problems like biomedical image analysis. The idea is to find centerlines of the desired structures in a given image while preserving the overall topology of the structure. In this proposal, we analyzed three works that are related to this problem. We presented the problems they solved, the methods they used, results of the works and how we can integrate these work to our research. The first paper focuses on an end-to-end 3D convolutional network to produce individual segments. The second paper is a survey paper on deformable models, their extensions and applications. The third paper is on a new method on unsupervised reconstruction of 3D point clouds. Finally, we talked about our current research, and future plans and discussed how to use these papers to improve our research to obtain topologically better capture of curvilinear structures.
Background papers
Flood-fliing Networks by Michal Januszewski, by Jeremy Maitin-Shepard, Peter Li, Jörgen Kornfeld, Winfried Denk and Viren Jain, CoRR 2016.
Deformable models in medical image analysis: a survey by Tim McInerney and Demetri Terzopoulus, Medical Image Analysis (MIA) 1996.
FoldingNet: Point Cloud Auto-Encoder via Deep Grid Deformation, by Yaoqing Yang, Chen Feng, Yiru Shen and Dong Tian, IEEE/CVF Conference on Computer Vision and Pattern Recognition 2017.
Practical information
- General public
- Free