摘要
从全局特征保持和局部特征保持的角度出发,提出一种稀疏近邻保持投影(SNPE)算法。该算法融合了稀疏重构信息和局部近邻重构信息。投影后的低维数据保持了高维数据的全局几何结构信息和局部近邻近似非线性的结构信息。在Yale、AR和UMIST上的实验表明所提算法是有效的。
In the term of global feature preserving and local feature preserving, a dimensionality reduction algorithm called sparse neighbourhood preserving projection （SNPE） is proposed. The algorithm fuses the sparse reconstruction information and local neighbourhood reconstruction information. The projected low-dimensional data preserve the global geometric structure information and local neighbourhood approximated nonlinear structure information of the high-dimensional data. Experiments operated on Yale, AR and UMIST face dataset show that the proposed algorithm is effective.
出处
《计算机应用与软件》
CSCD
北大核心
2013年第6期266-268,294,共4页
Computer Applications and Software
关键词
降维
稀疏保持投影
近邻保持嵌入
加权融合
平衡参数
Dimensionality reduction
Sparse preserving projection
Neighbourhood preserving embedding
Weighted fusion
Tradeoff parameter
