Wu, Shihao (2018). Geometric Structure Extraction and Reconstruction. (Thesis). Universität Bern, Bern
|
Text
Thesis_SH.compressed.pdf - Thesis Available under License BORIS Theses Standard License. Download (11MB) | Preview |
Abstract
Geometric structure extraction and reconstruction is a long-standing problem in research communities including computer graphics, computer vision, and machine learning. Within different communities, it can be interpreted as different subproblems such as skeleton extraction from the point cloud, surface reconstruction from multi-view images, or manifold learning from high dimensional data. All these subproblems are building blocks of many modern applications, such as scene reconstruction for AR/VR, object recognition for robotic vision and structural analysis for big data. Despite its importance, the extraction and reconstruction of a geometric structure from real-world data are ill-posed, where the main challenges lie in the incompleteness, noise, and inconsistency of the raw input data. To address these challenges, three studies are conducted in this thesis: i) a new point set representation for shape completion, ii) a structure-aware data consolidation method, and iii) a data-driven deep learning technique for multi-view consistency. In addition to theoretical contributions, the algorithms we proposed significantly improve the performance of several state-of-the-art geometric structure extraction and reconstruction approaches, validated by extensive experimental results.
Item Type: | Thesis |
---|---|
Dissertation Type: | Single |
Date of Defense: | 2018 |
Uncontrolled Keywords: | Points representation, meso-skeleton, consolidation, data consolidation, filtering, clustering, dimensionality reduction, manifold denoising, generative adversarial network, multi-view reconstruction, multi-view coherence, specular-to-diffuse, image translation |
Subjects: | 000 Computer science, knowledge & systems 500 Science > 510 Mathematics 000 Computer science, knowledge & systems > 070 News media, journalism & publishing |
Institute / Center: | 08 Faculty of Science > Institute of Computer Science (INF) |
Depositing User: | Admin importFromBoris |
Date Deposited: | 25 Jan 2019 12:56 |
Last Modified: | 08 Apr 2021 12:18 |
URI: | https://boristheses.unibe.ch/id/eprint/796 |
Actions (login required)
View Item |