超像素驱动的代理辅助多目标聚类图像分割  

A surrogate-assisted multi-objective clustering imagesegmentation motivated by superpixel

作　　者：赵凤 张莉阳 ZHAO Feng;ZHANG Liyang(School of Communication and Information Engineering,Xi'an University of Posts and Telecommunications,Xi'an 710121,China;Key Laboratory of Electronic Information Application Technology for Scene Investigation of Ministry of Public Security,Xi'an University of Posts and Telecommunications,Xi'an 710121,China)

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710121 [2]西安邮电大学电子信息现场勘验应用技术公安部重点实验室,陕西西安710121

出　　处：《西安邮电大学学报》2020年第6期31-37,共7页Journal of Xi’an University of Posts and Telecommunications

基　　金：国家自然科学基金项目(61571361,61102095,61671377);西安邮电大学“西邮新星”团队支持计划项目(xyt2016-01)。

摘　　要：为了提高多目标进化聚类算法的分割效果和时间效率,提出一种超像素驱动的代理辅助多目标聚类图像分割算法。利用超像素策略对图像进行预处理得到超像素区域,提取每个超像素区域的代表特征得到超像素信息,采用两个融合超像素信息的适应度函数。通过基于代理辅助参考向量引导的多目标进化算法优化这两个适应度函数得到最优解集,构造融合超像素信息的最优解评价指标得到最优解。实验结果表明,相较于其他多目标进化聚类算法,该算法的分割效果和时间效率均有所提高。In order to improve the segmentation effect and time efficiency of multi-objective evolutionary clustering algorithms,a surrogate-assisted multi-objective clustering image segmentation motivated by superpixel is proposed.Superpixel strategy is used to preprocess the image.The representative features of each superpixel area are extracted to obtain the superpixel information.The two fitness functions are adopted by introducing the superpixel information.An optimal solution set is then obtained by optimizing these two fitness functions based on the multi-objective evolutionary algorithm guided by the surrogate-assisted reference vectors.An optimal solution is finally selected by constructing the optimal solution evaluation index with superpixel information.Experimental results show the segmentation effect and time efficiency of this algorithm are improved compared to other multi-objective evolutionary clustering algorithms.

关 键 词：图像分割 多目标进化 模糊聚类 超像素 代理辅助优化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]