Rime Optimization Algorithm with Kapur's Entropy for Multilevel Segmentation of Retinal Images

Sajad Ahmad Rather, P. Shanthi Bala, Partha Pratim Roy

Abstract: Accurate image segmentation is fundamental for dividing images into distinct regions based on pixel intensity, enabling precise analysis and interpretation of their features. However, conventional segmentation methods often encounter challenges in complex pixel spaces, including local minima entrapment, premature convergence, and increasing computational demands as the number of threshold levels grows. To address these limitations, this study employs the Rime Optimization Algorithm (RIME), an advanced metaheuristic technique tailored for multilevel thresholding. Inspired by the natural formation of rime-ice, RIME integrates broad exploration through soft-rime and refined exploitation via hard-rime. The algorithm further boosts optimization through an improved greedy selection mechanism and dynamic population updates, ensuring proper balance between exploration and exploitation stages. Coupled with Kapur’s entropy, RIME enhances segmentation accuracy by efficiently determining optimal pixel thresholds in reduced computational time. The algorithm’s performance is comprehensively evaluated using retinal imaging datasets from Kaggle, with segmentation quality assessed through metrics such as Structural Similarity Index Measure (SSIM), Feature Similarity Index Measure (FSIM), Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), optimal thresholds, and fitness values. Statistical validation is performed using the Wilcoxon signed-rank and Friedman ranking tests. Experimental results demonstrate that RIME significantly outperforms eleven competitive algorithms, achieving an SSIM of 0.88, FSIM of 0.93, and PSNR of 29.70, highlighting its effectiveness in addressing complex image segmentation challenges.

External IDs:dblp:conf/acpr/RatherBR25