Multilevel Thresholding With Metaheuristic Methods

In this study, a multi-level thresholding method (2DYOH-PSO) based on 2D non-local means histogram is proposed, taking into account the fast convergence rate of the PSO method to reduce the computation time and improve the multi-level thresholding performance. The proposed 2DYOH-PSO method has been realized by using the two-dimensional Renyi's entropy-based thresholding method. Experimental studies are conducted for 300 images in the Berkeley-Benchmark dataset, taking into account different level threshold values. The performance of the proposed 2DYOH-PSO method is evaluated by comparing the existing 5 different threshold determination methods (Differential Evaluation, Artificial Bee Algorithm, Gravity Search Algorithm, Kbest Gravity Search Algorithm, and Chaotic Kbest Gravity Search Algorithm). The performance of the 2DYOH-PSO method is determined using 12 different performance evaluation indices. In the case of 3-level thresholding with 2DYOH-PSO in terms of 12 performance evaluation indexes with 5 different methods, the performance of the segmentation processes shows improvements such that 2.63% in BDE, 0.83% in PRI, 15.5% in SSIM, 13.2% in RMSE, 8.63% in PSNR, 35% in CC, 13,9% in AD, 14.75% in MD, 10.04% in NAE, respectively. In the case of 5-level thresholding with 2DYOH-PSO, the performance of the segmentation processes shows 1% improvement in BDI, 0,85% in FSIM, 15,35% in RMSE, 8,88% in PSNR, 0.85% in CC and 12.8% in AD with the experimental studies.