A novel equilibrium optimization algorithm for multi-thresholding image segmentation problems

Image segmentation is considered a crucial step required for image analysis and research. Many techniques have been proposed to resolve the existing problems and improve the quality of research, such as region-based, threshold-based, edge-based, and feature-based clustering in the literature. The researchers have moved toward using the threshold technique due to the ease of use for image segmentation. To find the optimal threshold value for a grayscale image, we improved and used a novel meta-heuristic equilibrium algorithm to resolve this scientific problem. Additionally, our improved algorithm has the ability to enhance the accuracy of the segmented image for research analysis with a significant threshold level. The performance of our algorithm is compared with seven other algorithms like whale optimization algorithm, bat algorithm, sine-cosine algorithm, salp swarm algorithm, Harris hawks algorithm, crow search algorithm, and particle swarm optimization. Based on a set of well-known test images taken from Berkeley Segmentation Dataset, the performance evaluation of our algorithm and well-known algorithms described above has been conducted and compared. According to the independent results and analysis of each algorithm, our algorithm can outperform all other algorithms in fitness values, peak signal-to-noise ratio metric, structured similarity index metric, maximum absolute error, and signal-to-noise ratio. However, our algorithm cannot outperform some algorithms in standard deviation values and central processing unit time with the large threshold levels observed.

Automated summary

Image segmentation is a crucial step in image analysis and research, with various techniques proposed to improve the quality of segmentation. A novel meta-heuristic equilibrium algorithm was introduced to find the optimal threshold for grayscale image segmentation, showing enhanced accuracy in research analysis. Comparative performance evaluation with other algorithms demonstrated superior results in fitness values, peak signal-to-noise ratio, structured similarity index, maximum absolute error, and signal-to-noise ratio.