A Research On Multi-Agent Evolutionary Algorithm For Image Segmentation

Image Segmentation is an important kind of image analysis technology. Image segmentation is a key procedure from image preprocessing to image analysis, thus playing an important role in image processing. There is no standard criteria for IS, different problems will be solved by different IS methods. Using evolutionary algorithms, especially genetic algorithms, to segment an image has been proved to be an efficient commonly used method with a good segmentation effect because of its robustness, parallelism and adaptability. However, evolutionary algorithms have some inherent defects such as high time-consuming and low efficiency when dealing with large scale of populations. Multi-agent evolutionary algorithm has the advantage of being asynchronous concurrent and its data distributed storage, which can compensate the disadvantages of evolutionary algorithm. Therefore, it’s a crucial to combine the evolutionary algorithm with multi-agent system for image processing.The theory basis of evolutionary algorithm and multi-agent genetic algorithm has been described in this paper, and the feasibility of the two algorithms for image segmentation is also studied. By using the cooperation among the populations in evolutionary algorithm, the organizational co-evolutionary algorithm for classification has been studied. Experiments show its good characteristic such as fast training and convergence rate, which validates the feasibility of the integration of the two algorithms.At last the hybrid algorithm, i.e. multi-agent genetic algorithm is applied to an image segmentation benchmark. The hybrid algorithm can segment images fast by improving individual cooperation ability and self-learning ability in genetic algorithm adding to obtain the entropy optimization. The feasibility and performance of the multi-agent genetic algorithm has been verified by the experiment on some image benchmarks. Compared to genetic algorithm, it can get a faster convergence rate and a better segmentation quality which can significantly improve the whole image segmentation performance.