Fast three-dimensional Otsu thresholding with shuffled frog-leaping algorithm

Three-dimensional (3-D) Otsu thresholding was regarded as an effective improvement over the original Otsu method, especially under low signal to noise ratio and poor contrast conditions. However, it is very time consuming for real-time applications. Shuffled frog-leaping algorithm (SFLA) is a newly developed memetic meta-heuristic evolutionary algorithm with good global search capability. In this paper, a fast threshold selection method based on SFLA is proposed to speed up the original 3-D Otsu thresholding for image segmentation. In this new paradigm, an updating rule is carefully designed to extend the length of each frog’s jump by emulating frog’s perception and action uncertainties. The modification widens the local search space thus helps to prevent premature convergence and improves the performance of the SFLA. It is then used to simplify the process for heuristic search of the optimal threshold instead of exhaustively exploring every possible threshold vector in three-dimensional space. Experimental results compared with the original 3-D Otsu and the fast recursive 3-D Otsu show that SFLA-based thresholding can exactly obtain the global optimal threshold with significant decrease in the computation time and the number of fitness function evaluation (FFE).