Image segmentation with pseudo branch and bound algorithm

• Published in: 2009 International Conference on Machine Learning and Cybernetics Vol. 4; pp. 2448 - 2452
• Main Authors: Hong-Gui Li, Xing-Guo Li
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2009
• Subjects: Branch and bound; branch and minicut; Cybernetics; grabcut; graph cuts; Image segmentation; Machine learning
• ISBN: 9781424437023; 1424437024
• ISSN: 2160-133X
• DOI: 10.1109/ICMLC.2009.5212215

Improved branch and minicut method for image segmentation is proposed. The most valuable contribution of proposed algorithm is that, accelerating branch and minicut method, avoiding exhaustive search on whole parameter space, and preserving segmentation quality at the same time, by loosening the condition of being a leaf node. Branch and minicut method utilizes branch and bound algorithm to explore the global minimum of energy function, and the lower bounds of branch and bound algorithm is quickly evaluated by graph cuts algorithm. Pseudo branch and bound algorithm is proposed, through relaxing the update condition of current optimal solution, which is equal to loosening the condition of being a leaf node in branch and minicut method. Experiments results of gray scale and color image segmentation show, proposed algorithm is faster than branch and minicut method, and has almost the same segmentation ability.