张咪咪1, 2,
刘汉强3
1.西安邮电大学通信与信息工程学院 ??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 ??西安 ??710121
3.陕西师范大学计算机科学学院 ??西安 ??710119
基金项目:国家自然科学基金(61571361, 61102095, 61671377)，西安邮电大学西邮新星团队(xyt2016-01)
详细信息
作者简介:赵凤：女，1980 年生，教授，研究方向为计算智能与图像处理
张咪咪：女，1992年生，硕士生，研究方向为图像处理
刘汉强：男，1981年生，副教授，研究方向为模式识别与图像处理
通讯作者:赵凤 fzhao.xupt@gmail.com
中图分类号:TP391计量
文章访问数:1136
HTML全文浏览量:486
PDF下载量:61
被引次数:0
出版历程
收稿日期:2018-06-20
修回日期:2018-12-14
网络出版日期:2019-01-18
刊出日期:2019-05-01
Multi-objective Evolutionary Semi-supervised Fuzzy Clustering Image Segmentation Motivated by Region Information
Feng ZHAO1, 2,,,Mimi ZHANG1, 2,
Hanqiang LIU3
1. School of Communication and Information Engineering, Xi’an University of Posts and Telecommunications, Xi’an 710121, China
2. 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
3. School of Computer Science, Shaanxi Normal University, Xi’an 710119, China
Funds:The National Natural Science Foundation of China (61571361, 61102095, 61671377), New Star Team of Xi’an University of Posts & Telecommunications (xyt2016-01)
摘要
摘要:现有的多目标进化聚类算法应用于图像分割时，往往是在图像像素层面上进行聚类，运行时间过长，而且忽略了图像区域信息使得图像分割效果不太理想。为了提高多目标进化聚类算法的分割效果和时间效率，该文将图像区域信息与部分监督信息引入多目标进化聚类，提出图像区域信息驱动的多目标进化半监督模糊聚类图像分割算法。该算法首先利用超像素策略获得图像的区域信息，然后结合部分监督信息，设计融合区域信息和监督信息的适应度函数，接着通过多目标进化策略对多个适应度函数进行优化得到最优解集。最后构造融合区域信息与监督信息的最优解评价指标，实现从最优解集中选取一个最优解。实验结果表明：与已有多目标进化聚类算法相比，该算法不但分割效果有所提升，而且运行效率得以提高。
关键词:图像分割/
多目标进化/
模糊聚类/
半监督聚类/
区域信息
Abstract:When multi-objective evolutionary clustering algorithms are applied to image segmentation, the image pixels are always utilized to be clustered. It results in a long running time. In addition, due to not considering the image region information, the image segmentation effect is not ideal. In order to improve the segmentation effect and time efficiency of the multi-objective evolutionary clustering algorithm, the image region information and some supervised information are introduced into multi-objective evolutionary clustering. Then a multi-objective evolutionary semi-supervised fuzzy clustering image segmentation algorithm driven by image region information is presented. First, the region information of the image is obtained through the super-pixel strategy. Second, two novel fitness functions are designed by introducing the supervised information and region information. Third, the multi-objective evolutionary strategy is used to optimize these two fitness functions to obtain an optimal solution set. Finally, an optimal solution evaluation index with region information and supervision information is constructed and utilized to select an optimal solution from the optimal solution set. Experimental results show the proposed algorithm outperforms comparison methods in segmentation performance and running efficiency.
Key words:Image segmentation/
Multi-objective evolutionary optimization/
Fuzzy clustering/
Semi-supervised clustering/
Region information
PDF全文下载地址:
https://jeit.ac.cn/article/exportPdf?id=dcb75239-a54d-4fe1-9e6a-ced135473315
