Abstract
To improve the quality and computational efficiency of image segmentation, and to overcome the limitations of the traditional K-means algorithm-such as sensitivity to initial cluster centers and susceptibility to local optima-this study proposes an Improved Dung Beetle Optimization (IDBO) algorithm and its application to K-means-based image segmentation. First, Latin Hypercube Sampling (LHS) is employed to initialize the population, enhancing diversity and uniformity in the search space and preventing premature convergence in early iterations. Second, a hybrid position updating strategy, combining a nonlinear decision factor with a competition mechanism, dynamically balances global exploration and local exploitation, improving adaptability across different search stages. Third, the Cauchy inverse cumulative distribution operator and tangent flight operator are integrated to perform dynamic perturbation and fine-tuning on optimal individuals, strengthening local exploitation and enhancing the ability to escape local optima. Comprehensive experiments on standard benchmark functions demonstrate that IDBO outperforms the original DBO and other comparative algorithms in convergence speed, optimization accuracy, and stability. The algorithm is further applied to optimize K-means clustering for image segmentation. Quantitative metrics, including Mean Squared Error (MSE) and Peak Signal-to-Noise Ratio (PSNR), confirm that IDBO-based segmentation achieves higher accuracy, better edge preservation, and improved texture fidelity. Additionally, an ablation study isolates the contributions of each enhancement strategy, demonstrating their complementary effects and validating the superiority of the integrated IDBO framework. These results highlight the potential of combining intelligent optimization and clustering algorithms to develop adaptive, high-performance image segmentation techniques.