摘要
针对子空间聚类应用中高达数以百万计信号的数据集合问题,为了实现快速聚类,提出了一种基于稀疏表示的概率子空间聚类算法。首先,每个信号由一个稀疏组合的基本元素(原子)表示,这些原子构成了字典矩阵的列;接着利用稀疏表示集推导出一个混合模式的原子和信号的共生矩阵;最后,通过共生矩阵的非负矩阵分解(NNMF)得到混合模式的组件,并根据最大似然(ML)准则估算每个信号的子空间。在YaleB人脸数据库上的实验结果表明,与其他几种最先进的方法相比,所提方法取得了较好的聚类精度。
To implement rapidly clustering for the very large signal collections problem in subspace clustering applications, a probabilistic subspace clustering algorithm based on sparse representations is proposed. Firstly, each signal is represented by a sparse combination of basis elements （ atoms）, which form the columns of a dictionary matrix. Then, the set of sparse representations is utilized to derive the co-occurrences matrix of atoms and sig- nals, which is modeled as emerging from a mixture model. Finally, the components of the mixture model are obtained via a non-negative matrix factori- zation （NNMF） of the matrix, and the subspace of each signal is estimated according to a maximum-likelihood （ML） criterion. Experi- mental results on YaleB face database show that proposed method has better clustering accuracies comparing with several latest approaches
出处
《电视技术》
北大核心
2014年第11期173-176,共4页
Video Engineering
基金
苏州市科技计划支撑项目(020142010)
关键词
人脸识别
稀疏表示
概率子空间聚类
字典学习
非负矩阵分解
face recognition
sparse representation
probabilistic subspace clustering
dictionary learning
non-negative matrix factorization
