代理多目标粒子群驱动的粗糙聚类图像分割算法

Surrogate multi-objective particle swam driven rough clustering for image segmentation

为了提高粗糙聚类算法应用于图像分割时的分割效果,提出一种代理多目标粒子群驱动的粗糙聚类图像分割算法。首先,通过自适应确定粗糙聚类上、下近似的阈值,减少人为干预;其次,利用粗糙聚类中边界样本占比构建动态惩罚因子,进而结合聚类的紧致性和可分性度量构造粗糙聚类目标函数,并联合聚类的连通性函数从不同角度共同评价聚类质量;最后,设计代理辅助的精英多目标粒子群优化策略,筛选精英粒子更新种群,得到最终的聚类中心,从而避免粗糙聚类算法对初始中心敏感和易陷入局部最优的问题并提升优化效率。实验结果表明:所设计的优化策略在标准测试问题上能够得到更好的优化结果;对比其他图像分割算法,该算法分割效果最佳。

In order to improve the segmentation effect of rough clustering algorithm when applied to image segmentation,a surrogate multi-objective particle swam driven rough clustering algorithm is proposed for image segmentation.First,an adaptive threshold determination mechanism is designed to determine the upper and lower approximations of rough clustering,and to reduce the manual intervention.Second,a penalty factor is designed by utilizing the proportion of samples in the boundary region in rough clustering,and then an effective rough clustering objective function is constructed with the combination of the compactness and the separation of clusters.The clustering quality is evaluated from different perspectives by the connectivity objective function of clusters.Finally,the surrogate-assisted elite multi-objective particle swarm optimization strategy is designed to screen the elite particles to update the population,and to obtain the final clustering center,thus avoiding the problem that the rough clustering algorithm is sensitive to the initial center and prone to fall into the local optimum,and improving the optimization efficiency.Experiment results show that the designed optimization strategy can get better results on the standard test problem;it has the best segmentation effect compared with other image segmentation algorithms.