Ketut Fundana received the PhD degree in computer science – Københavns Universitet

Datalogisk Institut, DIKU > Nyheder > DIKU-nyheder 2010 > Ketut Fundana received...

20. maj 2010

Ketut Fundana received the PhD degree in computer science

Ketut Fundana, was Thursday 20th May 2010 awarded the PhD degree in computer science for his thesis on Variational Segmentation Problems using Prior Knowledge in Imaging and Vision

The Assessment Committee, comprising as Chairman:  Associate professor Francois Lauze, Department of Computer Science, Copenhagen University (DIKU) and memhers: Professor Daniel Cremers, Department of Computer Science, Germany and Professor Ron Kimmel, Department of Computer Science, Technion Israel Institute of Technology unanimously, accepted the thesis.

DIKU contratulates Ketut Fundana with the PhD degree.

From the defence:




Image segmentation aims to partition a given image into semantically meaningful regions or objects with consistent properties. It is one of the most important problems in Computer Vision which provides crucial information for a variety of high level applications, such as 3D reconstruction, video surveillance, object recognition, etc.

Segmentation is a very difficult problem to solve with many challenges depending on the data and applications especially since the definition of regions or objects is subjective. Moreover, due to the presence of noise, clutter and occlusions, which commonly appear in images, the use of image information alone can lead to inconsistent segmentation results.  As in human vision which tends to integrate low-level and high-level information, most real-life applications need to have some prior knowledge about the regions or objects of interest beforehand. Due to the importance of the prior knowledge in the segmentation process, mathematical modeling of prior knowledge and its integration into segmentation models become critical.

We deal with variational segmentation problems using region-based active contour models, in both convex and non-convex formulations. The models are then coupled with prior knowledge such as smoothness terms, occlusion information and shape priors, in order to segment regions or objects of interest. This dissertation contains several contributions: a novel shape priors formulation and its shape gradient, a novel variational contour matching to detect and locate occlusions, shape submanifolds for segmentation and viewpoint tracking, novel shape priors formulations for convex segmentation models, a novel gradient descent procedure for pose parameters estimation, and the adaptation of convex segmentation models to manifolds. We demonstrate the robustness of the proposed models through a series of experiments.

For an electronic copy of the thesis, please contact Dina Riis Johannessen, 35 3214 23,