Abstract
Metaheuristic Algorithms (MAs) are commonly used in the scope of digital image processing, in particular, image segmentation processes. This is evident in Multilevel Thresholding (MTH) methods, where the optimal threshold configuration must be found to produce high-quality segmented images. Minimum Cross-Entropy (MCE) is one of the most prominent techniques for MTH due to its simplicity and efficiency. This article proposes a comparison of recent MAs that have not yet been implemented for image segmentation. Six recently published MAs were implemented and tested on nine complicated images selected from the BSDS300 dataset. Analyzing the results reveals the best algorithm when MCE is used as the objective function. Central tendency indicators, such as Standard Deviation and mean, are also used to analyze the five threshold values. Additionally, three quality indicators used in processing images are analyzed: Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM), and Feature Similarity (FSIM). The result of this analysis allows for the quality of the segmentation of each algorithm used in the comparison. The metrics with the highest values are indicative of the most effective algorithm in terms of segmentation performance.

This publication has 41 references indexed in Scilit: