.png){kind=icon}
返回
A Multilevel Thresholding Algorithm Based on Grey Wolf Optimization and Expectation Maximization
.png){kind=icon}
PRE
AI
DOI：10.5829/ije.2025.38.12c.13.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Multilevel image thresholding is essential for segmenting images. Expectation Maximization (EM) is effective for finding thresholds; but, it is sensitive to starting points. The Grey Wolf Optimizer (GWO) is fast at finding thresholds but can get stuck in local optima. This paper presents a new algorithm, EM+GWO, combining both methods to improve segmentation. EM estimates Gaussian Mixture Model (GMM) coefficients, while GWO finds better solutions when EM is stuck. GWO adjusts GMM parameters using Root Mean Square Error (RMSE) for the best fit. The algorithm was tested on nine standard images, evaluating global fitness, PSNR, SSIM, FSIM, and computational time. The results show that EM+GWO significantly enhances image segmentation effectiveness. Statistical tools indicate that RCG achieves the best RMSE and PSNR in 7 out of 9 test images, and it holds the highest rank in both SSIM and FSIM. The average execution time of each algorithm was calculated, showing that EM+GWO has an acceptable running time compared to EM and GWO. This balance between computational efficiency and improved segmentation performance makes the proposed EM+GWO algorithm a robust and effective solution for image segmentation tasks. Overall, the combination of EM and GWO methods provides a more reliable and accurate approach to optimizing image segmentation, avoiding local optima, and enhancing overall performance.
Keyword:
Image Segmentation
Multilevel Thresholding
Expectation Maximization
Grey Wolf Optimization
Gaussian Mixture Model
