Deformable graph models for image segmentation
Brief Description of Project:
Surface-based graph cut methods are an attractive alternative to deformable models for image segmentation because of their global optimality and their robustness to noise and leaking problems. However, while the graph-cut algorithms are guaranteed to find the minimum-cost surface within the given search space, their initialization can severely affect the shape and extent of the search space itself. For example, if a branch is missing from the initial surface, it cannot be recovered during the graph optimization because it is outside the search space. This is in contrast to deformable models, such as snakes, where the surface adapts to the underlying image data by deforming throughout the optimization process. The goal of this project is to develop a new algorithm that combines the advantages of the two paradigms: a deformable graph model that can deform like a snake to dynamically adapt its search space while maintaining its global minimum guarantee and thus robustness. I expect this project to lead to a submission to the SPIE Medical Imaging Conference by the end of the summer.
Nature of Supervision:
I will meet with the student weekly. I have additional meetings as needed (e.g., near conference deadlines, etc). Both me and the other senior lab members (grad students/postdocs) are available for more frequent help as needed.
A Brief Research Plan (period is for 10 weeks):
1 week - project overview and plan development
6 weeks - project implementation
1 week - statistical evaluation
2 week - project write-up
Number of Open Slots: 1
Name: Ipek Oguz
Department: Electrical Engineering and Computer Science