Abstract: Image segmentation is a key part in image processing field. Based on the entropic thresholding segmentation methods, a novel algorithm based on particle swarm optimization is presented in this paper. We also analyze the convergence of the new algorithm based on Bayes’s theorem and stochastic transform process. In experiments, benchmark image and SAR image are selected, and the algorithm runs ten times independently and the running time is selected as the evaluation of the algorithm complexity. It shows that the algorithm presented in this paper can find better solutions with much little complexity. Experiments results show that this method is feasible and effective

摘要： 图像分割是自动目标识别的关键和首要步骤。群智能作为一类新兴的演化计算技术已被越来越多的研究者关注。本研究将群智能中的粒子群优化算法应用到图像分割中，提出了一种新的图像分割算法。新方法基于最佳熵阈值分割技术，用粒子群优化算法自适应选取分割阈值，基于Bayes定理和随机状态转移过程对新算法收敛性的分析表明，新方法能以概率1找到图像的最佳熵阈值。在仿真实验中，针对基准图像和SAR图像分割问题,将遗传算法与粒子群优化算法分别独立运行10次，对10次得到的阈值以及均值、方差进行了比较，并将运行时间作为算法复杂度的评价指标。统计结果显示，本文算法不仅能够对图像进行准确的分割，而且运行时间明显较短。仿真结果表明，基于粒子群优化的图像分割算法是可行的、有效的。