Multilevel Thresholding Segmentation for Color Image Using Modified Moth-Flame Optimization

Abstract

Multilevel thresholding has got more attention in the field of image segmentation recently. However, it is still challenging and complicated for color image segmentation in many applications. To mitigate the above conditions, a novel multilevel thresholding algorithm consists of two innovative strategies is proposed on the basis of moth-flame optimization (MFO) to develop the SAMFO-TH algorithm. On one hand, a creative self-adaptive inertia weight scheme is used to enhance both the exploration and exploitation, on the other hand, a newly proposed thresholding (TH) heuristic is embedded into MFO to improve the global performance in multilevel thresholding. To find the optimal threshold values of an image, Otsu's variance, and Kapur's entropy criteria are employed as fitness functions. The experiments have been performed on ten color images including six natural images and four satellite images at different threshold levels with a comparison of other eight meta-heuristic algorithms: multi-verse optimizer (MVO), whale optimization algorithm (WOA), standard MFO, and so on. The experimental results are presented in terms of computational time (CPU time), mean value to reach (MVTR), standard deviation (STD), mean square error (MSE), peak signal to noise ratio (PSNR), structural similarity (SSIM), feature similarity (FSIM), probability rand index (PRI), the variation of information (VoI), and threshold value distortion (TVD). The results demonstrate that the proposed SAMFO-TH outperforms other competitive algorithms and has superiority concerning stability, accuracy, and convergence rate, which can be applied to practical engineering problems.