An improved opposition-based Runge Kutta optimizer for multilevel image thresholding

Minimum cross-entropy is widely used to find the best threshold values for image segmentation; this technique is known as MCET. However, when the number of thresholds increases, it becomes computationally expensive. Under such circumstances, employing a metaheuristic algorithm (MA) is a good choice. The Runge Kutta (RUN) optimization algorithm is MA recently proposed for solving global optimization problems. Like other MA, the RUN tends to fall into sub-optimal solutions presenting a premature convergence, especially in high-dimensional problems. This article aims to improve the RUN by merging it with opposition-based learning (OBL), creating a hybrid algorithm called RUN-OBL. By doing this hybridization, the RUN can search in two directions, and its performance is considerably improved, solving its drawbacks. One of the most significant improvements in this paper is that the RUN-OBL can handle high-dimensional search spaces escaping from local optimal solutions. The performance of the RUN-OBL is tested over different experiment series. Firstly, the RUN-OBL is tested using the CEC’17 set of benchmark functions for numerical optimization. Here, comparisons with other MA and the Friedman test were conducted. Secondly, the proposed algorithm segments digital images using the MCET. A set of benchmark images with different degrees of complexity are used. The RUN-OBL for image thresholding is tested in the third experiment series over a set of medical images (chest x-ray). A comparative study of the segmentation results is conducted to verify the efficiency of the proposal. Here different MA and other image segmentation methodologies were used. The comparisons were performed in terms of the fitness functions, peak signal-to-noise ratio (PSNR), feature similarity index (FSIM), and structural similarity index (SSIM). Experimental results demonstrate that the proposed RUB-OBL approach is better regarding quality and consistency for segmentation purposes and the optimization of complex problems.