Abstract
Image thresholding is one of the fastest and easiest approach for image segmentation and serves as a preprocessing step in computer vision and image processing applications, such as surveillance, image perception, scene understanding, artificial intelligence, augmented reality, biomedical imaging, remote sensing, image fusion, etc. Metaheuristic approaches have recently gained attention in the field of image segmentation. The standard COOT algorithm is a good alternative for solving complex optimization problems; however, it suffers from drawbacks, such as stagnation and insufficient balance between exploration and exploitation. This paper proposes the application of an improved COOT (ICOOT) optimization algorithm for multilevel image thresholding. In the proposed method, Lévy flights are incorporated to enhance the exploration capability of COOT, and quasi-opposition-based learning is introduced to improve the exploitation capacity and balance exploration and exploitation. To verify the efficiency of the ICOOT algorithm, it has been tested for solving complex optimization problems from the CEC'17 benchmark. The ICOOT algorithm is also employed to calculate the threshold values in image segmentation via Otsu's entropy as an objective function for practical purposes. In addition to testing its performance in image thresholding, the proposed ICOOT algorithm has also been tested on benchmark images and computed tomography (CT) images from COVID-19 patients. The presented approach is compared with various state-of-the-art algorithms, and the ICOOT results outperform them.