Meta-heuristic algorithms are among the technologies that have good performance in multilevel threshold image segmentation by obtaining optimal thresholds. However, most studies in the literature consider either a single objective function or images of a single type or low threshold levels, due to the drawbacks of poor ability to balance global and local search, premature convergence in high dimension, or low convergence efficiency of existing work in handling multi-task image segmentation. This paper aims to address these drawbacks and to develop search mechanisms and an enhanced optimizer for multilevel threshold image segmentation considering simultaneously different objective functions, both grayscale and color images, and both low and high threshold levels. More precisely, to improve the capability of balancing between global exploration and local exploitation, firstly a novel search mechanism ASSM inspired by the salp swarm optimization algorithm (SSA) is proposed, which is shown to have universality in improving a class of swarm intelligence optimization algorithms called DP-algorithms. Then, by proposing hierarchical vertical-horizontal search (HVHS) strategy and combining it with improved circle chaotic mapping initialization, lens opposition-based learning, and Lévy flight strategy, a multi-strategy collaborative ENCOA framework is constructed to prevent premature convergence in high-dimensional solution space. To evaluate the performance of the ENCOA, comparison experiments are implemented on CEC2017 benchmark suite and four engineering problems. Finally, the ENCOA is applied to multilevel threshold image segmentation on 6 grayscale images and 4 color images, by taking both Kapur’s entropy and Otsu between-class variance as the objective functions, and under threshold levels ranging from 4 to 32. It is shown that the ENCOA outperforms other recent-related algorithms in terms of both convergence accuracy and segmentation quality, especially when dealing with high threshold segmentation.
Liu et al. (Tue,) studied this question.