Unsupervised hierarchical image segmentation through fuzzy entropy maximization

The fuzzy c-partition entropy has been widely adopted as a global optimization technique for finding the optimal thresholds when performing multilevel gray image segmentation. Nevertheless, existing fuzzy c-partition entropy approaches generally have two limitations, i.e., partition number c needs to be manually tuned for different input and the methods can process grayscale images only. To address these two limitations, an unsupervised multilevel segmentation algorithm is presented in this paper. The core step of our algorithm is a bi-level segmentation operator, which uses binary graph cuts to maximize both fuzzy 2-partition entropy and segmentation smoothness. By iteratively performing this bi-level segmentation operator, multilevel image segmentation is achieved in a hierarchical manner: Starting from the input color image, our algorithm first picks the color channel that can best segment the image into two labels, and then iteratively selects channels to further split each labels until convergence. The experimental results demonstrate the presented hierarchical segmentation scheme can efficiently segment both grayscale and color images. Quantitative evaluations over classic gray images and the Berkeley Segmentation Database show that our method is comparable to the state-of-the-art multi-scale segmentation methods, yet has the advantage of being unsupervised, efficient, and easy to implement.