摘要
设计了一种基于快速谱聚类的图像分割算法,该算法利用余弦相似度构造相似度矩阵,避免了传统谱聚类算法中尺度因子的精确设置问题,提高了算法效率.在谱映射的过程中,该算法采用了Nystrm逼近策略,降低了谱聚类算法的复杂度和内存消耗.在Berkeley图像库上的图像分割实验证明了算法的有效性.
An image segmentation approach based on a fast spectral clustering algorithm is proposed, in which cosine similarity is used to attain similarity matrix. As a result, the problem of accurately setting the scale factor in the traditional spectral clustering algorithm is avoided, and the efficiency of the algorithm is improved. To efficiently apply the algorithm to image segmentation, Nystrom approximation strategy is used in the course of spectral mapping to reduce the computation complexity and memory consumption. Experimental results on Berkeley image database show the validitv of the algorithm.
出处
《应用科技》
CAS
2012年第2期26-30,共5页
Applied Science and Technology
基金
国家自然科学基金资助项目(60975042)
关键词
图像分割
谱聚类
余弦相似度
Nystrm逼近
image segmentation
spectral clustering
cosine similarity
Nystrom approximation
