Modified dragonfly algorithm based multilevel thresholding method for color images segmentation

作     者：Peng, Xiaoxu Jia, Heming Lang, Chunbo 

作者机构：Northeast Forestry Univ Coll Mech & Elect Engn Harbin 150040 Heilongjiang Peoples R China 

出 版 物：《MATHEMATICAL BIOSCIENCES AND ENGINEERING》 (Math. Biosci. Eng.)

年 卷 期：2019年第16卷第6期

页      面：6467-6511页

核心收录：

学科分类：0710[理学-生物学] 07[理学] 0701[理学-数学] 070101[理学-基础数学] 

基　　金：Fundamental Research Funds for the Central Universities [2572019BF04] National Nature Science Foundation of China Northeast Forestry University Horizontal Project [43217002, 43217005, 43219002] 

主　　题：Dragonfly algorithm multilevel thresholding Kapur's entropy minimum cross entropy Otsu method elite opposition-based learning differential evolution 

摘      要：Accurate image segmentation is the preprocessing step of image processing. Multi-level threshold segmentation has important research value in image segmentation, which can effectively solve the problem of region analysis of complex images, but the computational complexity increases accordingly. In order to overcome this problem, an modified Dragonfly algorithm (MDA) is proposed to determine the optimal combination of different levels of thresholds for color images. Chaotic mapping and elite opposition-based learning strategies (EOBL) are used to improve the randomness of the initial population. The hybrid algorithm of Dragonfly Algorithms (DA) and Differential Evolution (DE) is used to balance the two basic stages of optimization: exploration and development. Kapur entropy, minimum cross-entropy and Otsu method are used as fitness functions of image segmentation. The performance of 10 test color images is evaluated and compared with 9 different meta-heuristic algorithms. The results show that the color image segmentation method based on MDA is more effective and accurate than other competitors in average fitness value (AF), standard deviation (STD), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM) and feature similarity index (FSIM). Friedman test and Wilcoxon s rank sum test are also performed to assess the significant difference between the algorithms.