An Improved Teaching–Learning-Based Optimization for Multilevel Thresholding Image Segmentation

Abstract

Due to its successful application image processing or computer vision system, image segmentation plays a significant role and has become a hot research hotspot. In this paper, we propose an improved teaching–learning-based optimization (NFDR-TLBO) to segment grayscale images via multilevel thresholding. In the proposed teaching–learning-based optimization variant, the neighborhood topology is introduced into the original teaching–learning-based optimization algorithm to maintain the exploration ability of the population and the fitness-distance-ratio mechanism is introduced into the original teaching–learning-based optimization algorithm to improve its optimization performance on complex numerical optimization problems. Moreover, the experimental results on 18 typical benchmark functions with different characteristics verify the feasibility and effectiveness of the proposed algorithm. Furthermore, the proposed algorithm is used to optimize Kapur entropy function in order to find the optimal threshold for image segmentation. Finally, the experimental simulations on different benchmark images show that the proposed algorithm is effective and efficient in improving the image segmentation performance in terms of peak-signal-to-noise ratio, structure similarity index and feature similarity.