Multilevel thresholding of color images using globally informed artificial bee colony algorithm

Multilevel image thresholding presents a computational challenge as the number of thresholds increases, requiring efficient optimization techniques. The artificial bee colony (ABC) algorithm is a widely used metaheuristic for addressing this problem. Despite the good performance of the ABC algorithm, it struggles with an inadequate balance between discovering new solutions and refining existing ones. This paper presents the globally informed artificial bee colony (giABC), an enhanced ABC variant, proposed for multilevel color image thresholding. To overcome the limitations of the ABC algorithm, giABC introduces two novel mutation operators. In the employed phase, solutions are dynamically guided toward the mean of the current better solutions, ensuring a sustained balance between global exploration and local enhancement. In the onlooker phase, solutions are further refined by combining attraction to the global best solution with adaptation to promising solutions, significantly enhancing both convergence speed and solution quality. The proposed giABC, along with the ABC, its two variants and the chaotically-enhanced Rao algorithm, were tested on twelve color images from the Berkeley dataset using Otsu's objective function. Experimental results show that giABC outperforms the other metaheuristics in accuracy, robustness, peak signal-to-noise ratio and structural similarity index, with Wilcoxon signed-rank tests confirming its statistical significance.