Optimal multilevel thresholding using a hybrid ant colony system

Thresholding is an important technique for image segmentation, yet the challenge of automatic determination of an optimum threshold value still exists. Otsu's method has been extensively applied to real-world image segmentation, but its exhaustive search procedure has limited its application to multilevel thresholding. For this reason, this article aims at finding a more applicable and effective segmentation procedure – a hybrid optimization scheme based on an ant colony system (ACS) algorithm with Otsu's method. The properties of discriminate analysis in Otsu's method are to analyze the separability among gray levels in an image. The ACS–Otsu algorithm, a non-parametric and unsupervised method, is an extension of the applications of ant colony optimization with a proper design of hierarchical search range and local search for image segmentation. The proposed method is capable of automatically generating the lower and upper bounds of the search range for each threshold and finding the optimal number of thresholds in a very short period of time. The experimental results show that the ACS–Otsu algorithm efficiently speeds up Otsu's method to a great extent and preserves its robustness at multilevel thresholding.