Variational Image Segmentation and Curve Evolution on Natural Images

Presenter: Baris Sumengen
Date: 09/02/2004
Time: 3:00 PM
Location: Engineering I, Room 2162


This dissertation explores variational methods for segmenting natural images. In the first part, we propose new designs for edge-based variational segmentation methods. Starting with the Edgeflow technique, two edge-based variational methods-a curve evolution method and an anisotropic diffusion method-are proposed. Extensive tests on standard data sets show that our methods outperform the current state of the art. This work is then extended to multi-scale edge detection and image segmentation.

In the second part of the dissertation, we explore region-based variational segmentation. A new class of variational segmentation cost functions are proposed. These cost functions are based on pair-wise dissimilarities between pixels. We minimize these cost functions within the variational framework using curve evolution, and propose a new segmentation method-the graph partitioning active contours (GPAC). Efficient implementations of GPAC are proposed and shown to work well on natural images. Finally, we show an application in which we use GPAC for pruning categories in large image databases for image search and retrieval.


Baris Sumengen was born in Trabzon, Turkey in 1975. He received the B.S. degree in Electrical Engineering and Mathematics from Bogazici University, in 1998 and M.Sc. Degree from UC,Santa Barbara in 2000. He is currently pursuing the Ph.D. degree at University of California, Santa Barbara. He is interested in image analysis, image segmentation and knowledge discovery from image databases.

Departmental Host: Professor B. S. Manjunath