Regularizing variational methods for robust object boundary detection

• Main Author: Fang, Wen
• Other Authors: Chan Kap Luk
• Format: Theses and Dissertations
• Language: English
• Published: 2009
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/14582
• Institution: Nanyang Technological University

In this thesis, robust object boundary detection by the variational methods is studied. The variational methods fall into two categories: boundary-based and region based. This thesis focuses on the boundary-based variational methods. However, as the boundary-based variational methods depend only on the boundary information to locate the object, they are very sensitive to the disruptions to the object boundary. If the object is partially occluded by other objects or is interfered by cluttered background, the evolving curve of the variational methods may be pulled away from the real boundary and converge to the wrong place. This is regarded as the “missing boundary” problem in this thesis. In order to achieve a robust object detection, additional boundary constraints need to be incorporated into the variational methods to regularize the curve evolution. In this thesis, two new boundary constraints are proposed. The first method incorporates the temporal information into the variational methods, while the second incorporates the shape prior information.