摘要
针对传统超像素生成方法的性能受超像素数目制约,提出一种互惠最近邻区域聚类生成超像素的方法。该方法利用k-means算法将图像细分为大量均匀同质类似单元格的小区域,使用互惠最近邻聚类算法以区域面积为约束条件在小范围搜寻互惠最近邻进行合并,从而生成超像素。通过实验与SLIC(simple linear iterative clustering)比较,结果表明,互惠最近邻区域聚类生成的超像素在较少的超像素数目情况下仍然具有高边界查全率和低欠分割错误的优点,可保持良好的分割质量。
In order to solve the issue that the performance of traditional methods of generating superpixels is restricted by the number of superpixels, we present a method for generating superpixels by clustering reciprocal nearest neighbors （RNN） of re- gions. Firstly, image is segmented into a large number of regular homogeneous small regions which are similar to cells by virtue of k means. Secondly, regions in a small distance with the regional area as constraint conditions are merged by RNN clustering to generate superpixels. To validate the effectiveness of the proposed method, results of experiment on BSDS 500 dataset of natural images show that this method for generating superpixels has advantages of high boundary recall and low under-segmentation error over SLIC superpixels, and maintains good segmentation quality.
出处
《中国科技论文》
CAS
北大核心
2013年第4期330-333,349,共5页
China Sciencepaper
基金
四川省教育厅青年基金资助项目(07ZB102)
关键词
超像素
互惠最近邻
K均值聚类
superpixels
reciprocal nearest neighbors
k-means clustering
