Multilevel thresholding based on Chaotic Darwinian Particle Swarm Optimization for segmentation of satellite images

• Published in: Applied soft computing 2017-06, Vol.55, p.503-522
• Main Authors: Suresh, Shilpa, Lal, Shyam
• Format: Article
• Language: English
• Subjects: Chaotic sequence; Convergence rate; Metaheuristic algorithms; Minimum cross entropy; Tsallis entropy
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2017.02.005

•This paper introduces an improved variant of Darwinian PSO algorithm based on chaotic functions.•It replaces random sequences by chaotic sequences mitigating the problem of premature convergence.•Efficiency of 10 defined chaotic maps are investigated and the best one was chosen.•The proposed algorithm is compared with five different chaotic variants of existing optimization algorithms.•It provides better convergence characteristics and segmentation results as compared with existing algorithms. This paper proposes an improved variant of Darwinian Particle Swarm Optimization algorithm based on chaotic functions. Most of the evolutionary algorithms faces the problem of getting trapped in local optima in its search for global optimum solutions. This is highly influenced by the use of random sequences by different operators in these algorithms along their run. The proposed algorithm replaces random sequences by chaotic sequences mitigating the problem of premature convergence. Experiments were conducted to investigate the efficiency of 10 defined chaotic maps and the best one was chosen. Performance of the proposed Chaotic Darwinian Particle Swarm Optimization (CDPSO) algorithm is compared with chaotic variants of optimization algorithms like Cuckoo Search, Harmony Search, Differential Evolution and Particle Swarm Optimization exploiting the chosen optimal chaotic map. Various histogram thresholding measures like minimum cross entropy and Tsallis entropy were used as objective functions and implemented for satellite image segmentation scenario. The experimental results are validated qualitatively and quantitatively by evaluating the mean, standard deviation of the fitness values, PSNR, MSE, SSIM and the total time required for the execution of each optimization algorithm.