Semnan University International Journal of Nonlinear Analysis and Applications 2008-6822 2026 04 26 An improved grasshopper optimization algorithm for multilevel thresholding image segmentation 10591 10.22075/ijnaa.2025.33478.5013 EN Leila Amiri Deartment of Computer Engineering and Information Technology, Razi University, Kermanshah, Iran Abdolah Chalechale Department of Computer Engineering and Information Technology, Razi University, Kermanshah, Iran Maryam Taghizadeh Faculty of Information Technology, Kermanshah University of Technology, Kermanshah, Iran Journal Article 2024 03 19 Multilevel thresholding is one of the most common, straightforward, and effective image segmentation algorithms. The most important issue in this method is choosing an appropriate threshold value. In such a way that by defining worthy thresholds, the image can be more accurately segmented. The Otsu approach is suitable for establishing the thresholds at two levels, but as the number of thresholds increases, the performance of Otsu diminishes in terms of time and segmentation accuracy. On the other hand, optimization techniques can be effective to address these challenges. As a result, it is used with optimization techniques to improve time and segmentation accuracy. In this paper, we propose an improved grasshopper optimization approach to enhance the quality of the segmented image and its accuracy. In the proposed method, multilevel thresholding image segmentation is performed by employing the Otsu method as an objective function. This research aims to enhance the grasshopper algorithm to improve image segmentation outcomes. For this purpose, various modifications were applied to the grasshopper method. The proposed algorithm is evaluated on some known images and compared with several optimization algorithms. The resultant modified grasshopper method outperforms other evolutionary algorithms like Whale, Firefly, and Artificial Bee Colony (ABC) optimization algorithms. The proposed IGOA algorithm outperforms other approaches at PSNR metric for threshold levels of 32 and 64 for 87.5% and 100% of images, respectively. Additionally, at SSIM metric, for both threshold levels of 32 and 64, it overcomes other approaches for 100% of images. https://ijnaa.semnan.ac.ir/article_10591_97aa31b6142e6843624454031c403b5c.pdf