基于改进粒子群算法的多阈值灰度图像分割

Multilevel thresholding gray-scale image segmentation based on improved particle swarm optimization

针对粒子群协同学习优化算法和粒子群综合性学习优化算法中的粒子更新规则不灵活问题,提出了一种新的粒子群多阈值灰度图像分割算法。该算法中的粒子更新策略能够根据粒子状态随时改变:迭代前期,粒子速度会不断增加以便加快搜索最优解;迭代后期,粒子速度开始变慢以便搜索更广区域,避免陷入局部最优;当粒子陷入局部最优时,让该粒子根据选出的榜样粒子学习,以便逃出局部最优。另外评价粒子最优解的目标函数采用的是图像指数熵。仿真实验结果表明改进的粒子群阈值优化算法在单阈值和多阈值情况下解决了传统熵算法执行效率低和粒子群优化算法更新规则不灵活易于陷入局部最优问题,分割结果非常好,而且稳定、高效。

Concerning the not flexible update rules of the cooperative learning Particle Swarm Optimization （PSO） and comprehensive PSO, a new algorithm based on improved PSO for gray-sale image segmentation using multilevel thresholding was proposed. The update strategy of particles depends on the state of the particle in that algorithm： in the early iterations, the velocity of particle is increasing in order to speed up the search for optimal solution; later, the velocity of particle begins to decrease so that the particle can search for a broader area, avoiding to trap in local optimization; when particles was caught in local optimization, the particle begins to update its＇ velocity and position according to the chosen example particle that can make the particle escape from the local optimization; in addition, the object function of evaluating every particle＇s optimal solution is image exponential entropy. The simulation experiment results show that the new improved PSO solves the problems of the low execution efficiency of the traditional entropy algorithm and the not flexible update rules and easily falling into local optimization of the PSO, and its segmentation results are very good, stable and efficient in the case of single thresholding and multilevel thresholding.