Research On Image Clustering Based On Differential Evolution

Evolutionary algorithm(EAs) is inspired by the wisdom of nature. EAs is designed new global optimization method that imitates biosphere evolutionary process. EAs doesn’t restricted by the specific characteristics of problems. The advantage of EAs is common, simple and parallel processing. EAs receives a lot of atteniion for it solve complex optimization problems, and it contain a wide range of algorithms, such as the Genetic algorithm, Particle swarm optimization, Differential evolution, etc.Global optimization is an attractive research area in the computer science and there are a considerable number of real-world decision processes that require the solving of global optimization problems. Differential evolution(DE) has become one of the most frequently used evolutionary algorithms for solving the continuous global optimization problems in recent years. The algorithm is encoded using real number, and swarm intelligent model is used for achieving the search for the solution.Like other EAs, DE is also a population based stochastic search and contains three main steps, mutation,crossover and selection.In this paper, we analysis DE and propose improvement strategies, then apply it to image clustering problem, and gets good results. Specific work include the following:(1)Based on the concept of center-based sampling, we proposed random-based sampling. When they investigated the closeness of points in a search space from an unknown solution, they found that the center of the search space has a great probability close to an unknown solution. Random-based sampling is the extension of center-based sampling and random-based sampling expands the search area of DE. Compared with center-based sampling, random-based sampling is more flexible and effective.(2)The DE algorithm is good at exploring the search space; however, it may be slow at exploitation of the solutions in the current population. In order to improve the performance of DE, this paper attempts to improve DE’s exploitation ability by integrating the one-step K-means clustering algorithm, random-based sampling and Gaussian sampling. The proposed DE algorithm is named clustering-based differential evolution with random-based sampling and Gaussian sampling(GRCDE). This paper compares other state-of-the-art evolutionary algorithms with the proposed algorithm. All experiments results indicate the Gaussian sampling and the random-based sampling remarkably accelerate the convergence rate and improves exploitation ability of DE. It shows that GRCDE has a better solving ability and our proposed GRCDE performs better than some DE variants.(3)Improve differential evolution for solving image clustering problem. Design the evaluation function based on image clustering. Through the experiment, in the applications of image clustering, this paper analyzes base on differential evolution for solving image clustering problem. From the results, it can be seen that GRCDE is significantly better than DE.