摘要
提出一种融合多尺度边缘流和归一化割的图像分割方法。首先采用多尺度边缘流,检测在大尺度和小尺度上均存在的边缘,并在小尺度上定位边缘,再通过边缘连接,得到封闭的边缘;然后把多尺度边缘流分割得到的区域作为归一化割中图的顶点,区域间的相似性度量作为相似性矩阵的元素,大大降低了归一化割对内存的需求,提高了特征方程的计算效率,使归一化割得到实用,同时也降低了边缘流的过分割现象。实验表明,在岩心扫描图像的分割中,该方法能得到很好的效果。
An image segmentation method which combines Multi-scale Edgeflow and Ncut（Normalized cut） was put forward.First,Multi-scale Edgeflow was used to detect the edges existing in both largescale and small scale,edge position was made in small-scale,then a closed edge was acquired through a marginal connection.Second,the region acquired by Multi-scale Edgeflow was regarded as the vertices of the graph in Ncut;the similarity criterion among regions was regarded as elements of similarity matrix.Thus,the memory requirement for Ncut was greatly reduced,and the calculating efficiency of the characteristic equation was enhanced.As a result,Ncut was put into practicability;over-segmentation of edgeflow was reduced at the same time.In the core scanning image segmentation,the experimental results show that it is a very good segmentation method.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2010年第3期461-464,469,共5页
Journal of Optoelectronics·Laser
基金
四川省科技攻关项目05GG021-026-03
关键词
融合
岩心图像
多尺度边缘流
归一化割
过分割
integration
core image
multi-scale edgeflow
normalized cut
over-segmentation
