An experimental comparison of modern methods of segmentation

• Published in: 2010 IEEE 8th International Symposium on Applied Machine Intelligence and Informatics (SAMI) pp. 247 - 252
• Main Authors: Karch, P., Zolotova, I.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-01-2010
• Subjects: Active contours; Artificial intelligence; Boundary conditions; Computational complexity; Cybernetics; Image segmentation; Informatics; Level set; Machine intelligence; Object detection
• ISBN: 1424464226; 9781424464227
• DOI: 10.1109/SAMI.2010.5423728

This paper deals with experimental comparison of classical method of Graph cut segmentation with segmentation using Active contours. From methods of Active contours are chosen to comparison two methods. First method is based on classic Active contour method and second method is based on Active contours independent on gradient of the edges. Application of Graph cut segmentation allows finding the optimal global segmentation with the best balance between regional and boundary conditions among all possible segmentations at met condition limitations. Active contour segments the image by iterative deformation contour till this contour divided the image on the regions. Active contours are often implemented with level set methods for their universality and performance, but disadvantage is their computational complexity. The second method of the active contour allows to detect objects whose boundaries are not necessarily defined by gradient. The end rule in this case does not depend on gradient of the image, as in classical model of active contour, but instead refer to a particular segmentation of the image.