Graph Degree Linkage: Agglomerative Clustering on a Directed Graph
被引:0
作者:
Zhang, Wei
论文数: 0引用数: 0
h-index: 0
机构:
Chinese Univ Hong Kong, Dept Informat Engn, Hong Kong, Hong Kong, Peoples R ChinaChinese Univ Hong Kong, Dept Informat Engn, Hong Kong, Hong Kong, Peoples R China
Zhang, Wei
[1
]
Wang, Xiaogang
论文数: 0引用数: 0
h-index: 0
机构:
Chinese Univ Hong Kong, Dept Elect Engn, Hong Hom, Hong Kong, Peoples R China
Chinese Acad Sci, Shenzhen Inst Adv Technol, Shenzhen, Peoples R ChinaChinese Univ Hong Kong, Dept Informat Engn, Hong Kong, Hong Kong, Peoples R China
Wang, Xiaogang
[2
,3
]
Zhao, Deli
论文数: 0引用数: 0
h-index: 0
机构:
Chinese Univ Hong Kong, Dept Informat Engn, Hong Kong, Hong Kong, Peoples R ChinaChinese Univ Hong Kong, Dept Informat Engn, Hong Kong, Hong Kong, Peoples R China
Zhao, Deli
[1
]
Tang, Xiaoou
论文数: 0引用数: 0
h-index: 0
机构:
Chinese Univ Hong Kong, Dept Informat Engn, Hong Kong, Hong Kong, Peoples R China
Chinese Acad Sci, Shenzhen Inst Adv Technol, Shenzhen, Peoples R ChinaChinese Univ Hong Kong, Dept Informat Engn, Hong Kong, Hong Kong, Peoples R China
Tang, Xiaoou
[1
,3
]
机构:
[1] Chinese Univ Hong Kong, Dept Informat Engn, Hong Kong, Hong Kong, Peoples R China
[2] Chinese Univ Hong Kong, Dept Elect Engn, Hong Hom, Hong Kong, Peoples R China
[3] Chinese Acad Sci, Shenzhen Inst Adv Technol, Shenzhen, Peoples R China
来源:
COMPUTER VISION - ECCV 2012, PT I
|
2012年
/
7572卷
基金:
中国国家自然科学基金;
关键词:
IMAGE SEGMENTATION;
D O I:
暂无
中图分类号:
TP18 [人工智能理论];
学科分类号:
081104 ;
0812 ;
0835 ;
1405 ;
摘要:
This paper proposes a simple but effective graph-based agglomerative algorithm, for clustering high-dimensional data. We explore the different roles of two fundamental concepts in graph theory, indegree and outdegree, in the context of clustering. The average indegree reflects the density near a sample, and the average outdegree characterizes the local geometry around a sample. Based on such insights, we define the affinity measure of clusters via the product of average indegree and average outdegree. The product-based affinity makes our algorithm robust to noise. The algorithm has three main advantages: good performance, easy implementation, and high computational efficiency. We test the algorithm on two fundamental computer vision problems: image clustering and object matching. Extensive experiments demonstrate that it outperforms the state-of-the-arts in both applications.