摘要
针对聚类算法在图像分割上存在分割效果和时间效率上的不足,基于网格聚类算法ShrinClus,提出一种新的图像分割方法,该方法通过把图像的RGB空间分割成网格,将所有像素点分配到原子网格当中,然后对非空原子网格集合进行收缩聚类,通过查找低密度的边缘网格来确定簇的边界,最终将原子网格的分类结果映射至像素点.该方法能有效地分割在RGB空间中存在部分重叠的图像,算法具有接近线性的时间复杂度.最后通过实验验证了新方法的有效性.
A new image segmentation method called SCIS, based on the grid-based clustering algorithm ＂ShrinClus＂, is proposed. In SCIS, the RGB space of image is partitioned into non-overlapping grids, and the pixel points then are allocated to atom grids. After the shrinking of all non-empty atom grids, the border of cluster is determined by searching for the marginal units, which are of low-density. SCIS can efficiently segment the image which has overlap in the RGB space. Experiments on image segmentation are given to illustrate the performance of the new method SCIS..
出处
《郑州大学学报（理学版）》
CAS
2007年第4期146-149,共4页
Journal of Zhengzhou University:Natural Science Edition
基金
厦门大学科研基金资助项目
编号0630-X01117
关键词
聚类分析
网格聚类
数据点收缩
图像分割
clustering analysis
grid-based clustering
shrinking point
image segmentation
