Multilevel Thresholding Image Segmentation Based on Multi-swarm Particle Swarm optimization with a Dynamic Learning Strategy and Kapur’s entropy

In this paper, the multilevel thresholding method based on Kapur’s entropy and the recently proposed multi-swarm particle swarm optimization with a dynamic learning strategy is considered. The multilevel thresholding method is extensively evaluated on ten benchmark images. The experimental results, which include the mean and standard deviation of Kapur’s entropy, obtained from forty independent executions of the thresholding method for each test image and the considered total number of thresholds, demonstrate that multi-swarm particle swarm optimization with dynamic learning strategy can be successfully applied to solve the multilevel thresholding problem.

I. Introduction

Image segmentation has an important role in image processing, computer vision and pattern recognition. Over the years, various approaches to image segmentation have been developed, such as region-based segmentation, edge-based segmentation, clustering-based segmentation, graph-based and thresholding [1]. Due to its effectiveness and simplicity, the thresholding approach to image segmentation has been attracting significant research attention [1, 2]. Over the years, a variety of metaheuristic algorithms have been applied for image segmentation based on multilevel thresholding. Some recent examples of metaheuristic algorithms used for image segmentation include the bacterial foraging algorithm [3], krill herd algorithm [1], modified grasshopper algorithm [2], Rao algorithms [4], evolutionary arithmetic optimization algorithm [5], adaptive equilibrium optimizer [6], whale optimization algorithm [7], chef-based optimization algorithm [8], chimp optimization algorithm [9], emperor penguin optimization [10].