A new social and momentum component adaptive PSO algorithm for image segmentation

In this paper, we present a new variant of Particle Swarm Optimization (PSO) for image segmentation using optimal multi-level thresholding. Some objective functions which are very efficient for bi-level thresholding purpose are not suitable for multi-level thresholding due to the exponential growth of computational complexity. The present paper also proposes an iterative scheme that is practically more suitable for obtaining initial values of candidate multilevel thresholds. This self iterative scheme is proposed to find the suitable number of thresholds that should be used to segment an image. This iterative scheme is based on the well known Otsu's method, which shows a linear growth of computational complexity. The thresholds resulting from the iterative scheme are taken as initial thresholds and the particles are created randomly around these thresholds, for the proposed PSO variant. The proposed PSO algorithm makes a new contribution in adapting 'social' and 'momentum' components of the velocity equation for particle move updates. The proposed segmentation method is employed for four benchmark images and the performances obtained outperform results obtained with well known methods, like Gaussian-smoothing method (Lim, Y. K., & Lee, S. U. (1990). On the color image segmentation algorithm based on the thresholding and the fuzzy c-means techniques. Pattern Recognition, 23, 935-952; Tsai, D. M. (1995). A fast thresholding selection procedure for multimodal and unimodal histograms. Pattern Recognition Letters, 16, 653-666), Symmetry-duality method (Yin, P. Y., & Chen, L H. (1993). New method for multilevel thresholding using the symmetry and duality of the histogram. Journal of Electronics and Imaging, 2, 337-344), GA-based algorithm (Yin, P. -Y. (1999). A fast scheme for optimal thresholding using genetic algorithms. Signal Processing, 72, 85-95) and the basic PSO variant employing linearly decreasing inertia weight factor. (C) 2010 Elsevier Ltd. All rights reserved.