Optimized K-means algorithm for image segmentation based on improved dung beetle algorithm.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Optimization Algorithms

Background:

• Traditional K-means clustering for image segmentation suffers from sensitivity to initial centers and local optima.
• Existing optimization algorithms may lack efficiency and accuracy in complex segmentation tasks.

Purpose of the Study:

• To propose an Improved Dung Beetle Optimization (IDBO) algorithm for enhanced K-means image segmentation.
• To improve segmentation quality, computational efficiency, and overcome limitations of traditional K-means.

Main Methods:

• Implemented IDBO with Latin Hypercube Sampling (LHS) for population initialization.
• Utilized a hybrid position updating strategy balancing global exploration and local exploitation.
• Integrated Cauchy inverse cumulative distribution and tangent flight operators for dynamic perturbation.

Main Results:

• IDBO demonstrated superior convergence speed, accuracy, and stability on benchmark functions compared to DBO and other algorithms.
• IDBO-optimized K-means segmentation achieved higher accuracy, better edge preservation, and improved texture fidelity (validated by MSE and PSNR).
• Ablation studies confirmed the effectiveness of individual enhancement strategies within the IDBO framework.

Conclusions:

• The proposed IDBO algorithm significantly enhances K-means clustering for image segmentation.
• IDBO offers a robust and adaptive approach for high-performance image segmentation.
• Combining intelligent optimization with clustering presents a promising direction for advanced image analysis techniques.