摘要
针对复杂结构图像中形态滤波的单一属性难以判定最大树节点状态的问题,提出了基于多变量属性分类的最大树图像形态滤波方法。首先标记图像的各个连通区域,将图像转换为最大树数据结构,然后计算最大树各个节点的面积、灰度值及Zernike矩属性值,并构成节点的属性向量,运用属性样本数据对支持向量机进行训练,获得支持向量机分类模型,最后根据多变量属性分类结果给出节点的枝剪策略。实验结果表明,该方法能有效地滤除复杂结构图像中不同灰度级、大小及形状的噪声区域,同时保留图像目标区域的细节特征。
It is difficult to delete or retain max-tree node by using single attribute in morphology filtering for complex content image. A novel pruning strategy based on the multivariate attribute classification rule is presented. Firstly, each connected region is labeled, and the image is transformed into the max-tree data structure. Then, some attribute values such as node area, gray value and first Zeruike moment are calculated. These separated node attribute values are assembled to form an attribute vector. Meanwhile, support vector ma- chine （SVM） is trained by utilizing a large number of attribute sample data, which can obtain a SVM classification model. Finally, the node state is judged by the multivariate attribute classification rule. The experimental results show that this method can not only effective- ly filter noise while preserving image detail, but also achieved less structural similarity index than other methods.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2015年第8期1735-1743,共9页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(61304253
61403426
61471170)
教育部博士点新教师基金(20130162120018)
湖南省自然科学基金(13JJ3111)
湖南省教育厅重点项目(14A078
13A048
15A100)项目资助
关键词
形态滤波
连通区域
最大树
枝剪策略
多变量属性分类
morphological filtering
connected region
max-tree
pruning strategy
multivariate attributes classification
