A multi-leader whale optimization algorithm for global optimization and image segmentation

Abstract

In this paper, a multilevel thresholding image segmentation method base on the enhancement of the performance of the whale optimization algorithm (WOA). The developed method, called the multi-leader whale optimization algorithm (MLWOA), aims to avoid the limitations of traditional WOA during the searching process, such as stagnation at the local optimum. This was achieved by integrating the different tools with WOA, such as memory mechanism, multi-leader method, self-learning strategy, and levy flight method. Each of these techniques has its own task, for example, the memory structure of traditional WOA and add a multi-leader mechanism to enhance the ability of exploration. The superiority of leaders will make more influence in MLWOA by adding a self-learning strategy. Also, it used levy flight trajectory to make the algorithm more robust and avoid premature convergence. To evaluate the performance of the developed MLWOA, a set of experiments are conducted using the CEC2017 benchmark. In addition, it is applied to determine the optimal threshold values to segment a set of images using the Otsu method, fuzzy entropy, and Kapur's entropy as a fitness function. The results of MLWOA are compared with well-known meta-heuristic algorithms inside the experiments. The comparison results indicated that MLWOA provides better performance in CEC2017 benchmark functions and shows high superiority in image segmentation in terms of performance measures. In addition, the MLWOA provides better results using Otsu, followed by the Fuzzy entropy and Kapur in terms of PSNR. In terms of SSIM, fuzzy entropy and Otsu have nearly the same SSIM value, but the fuzzy entropy provides better results.