Image Segmentation using Curve Evolution

Baris Sumengen, B. S. Manjunath, and Charles Kenney
Department of Electrical and Computer Engineering
University of California, Santa Barbara, CA 93106-9560
{sumengen, manj, kenney} [at] ece.ucsb.edu

Abstract

A novel scheme for image segmentation is presented. The technique is based on the integration of ideas from geodesic active contours and a recently proposed edgeflow segmentation. Given an image a 2-D vector is constructed at each pixel location. This vector points in the direction of potential boundary pixels. The computation of the 2-D vector field is based on image intensity, color and texture gradients. Following this, an initial curve is instantiated and propagated to separate the image into foreground and background regions. The curve propagation is guided by the above mentioned vector field. The proposed approach thus utilizes an edge-based segmentation method and extends traditional PDE based curve evolution methods to texture image segmentation, and avoids the post-processing problems in edge linking and boundary detection.

[PDF] [BibTex] See also: #206
B. Sumengen, B. S. Manjunath, C. Kenney,
Asilomar Conference on Signals, Systems and Computers, vol. 2, pp. 1141-1145, Monterey, CA, Nov. 2001.
Node ID: 338 , DB ID: 136 , VRLID: 94 , Lab: VRL , Target: Conference
Subject: [Detection on Images and Videos] « Look up more