Effective Image Clustering with Differential Evolution Technique

Abstract

The paper presents a novel approach of clustering image datasets with differential evolution (DE) technique. The differential evolution is a parallel direct search population based optimization method. From our simulations it is found that DE is able to optimize the quality measures of clusters of image datasets. To claim the superiority of DE based clustering we have compared the outcomes of DE with the classical K-means and popular Particle Swarm Optimization (PSO) algorithms for the same datasets. The comparisons results reveal the suitability of DE for image clustering in all image datasets.

A K-Means Clustering Algorithm Based on Enhanced Differential Evolution

“…The parameter settings are configured as: The population size is 10 times of the dimension of data set, and the maximum number of iterations G max is 50. In DE-kmeans algorithm [6][7], the scale factor F is 0.5, and the crossover probability CR is 0.1. In this paper, the scale factor F is a Laplace random number, and the crossover probability CR is self-adaptive.…”

“…This guides the search towards an optimal solution for achieving global optimization. The k-means algorithm has been improved by using differential evolution as reported in literatures [6] [7]. It has been shown that differential evolution results in better performance, compared with the k-means algorithms using other evolutionary algorithms like the traditional genetic or particle swarm optimization.…”