Issue No. 04 - April (2006 vol. 28)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/TPAMI.2006.76
Baris Sumengen , IEEE Computer Society
B.S. Manjunath , IEEE
In this paper, we introduce new types of variational segmentation cost functions and associated active contour methods that are based on pairwise similarities or dissimilarities of the pixels. As a solution to a minimization problem, we introduce a new curve evolution framework, the graph partitioning active contours (GPAC). Using global features, our curve evolution is able to produce results close to the ideal minimization of such cost functions. New and efficient implementation techniques are also introduced in this paper. Our experiments show that GPAC solution is effective on natural images and computationally efficient. Experiments on gray-scale, color, and texture images show promising segmentation results.
Curve evolution, active contours, image segmentation, pairwise similarity measures, graph partitioning.
B. Manjunath and B. Sumengen, "Graph Partitioning Active Contours (GPAC) for Image Segmentation," in IEEE Transactions on Pattern Analysis & Machine Intelligence, vol. 28, no. , pp. 509-521, 2006.