Abstract:

Aiming at the problem of Locality Preserving Projection(LPP) does not provide the discriminating information of data set, this paper proposes an algorithm named Extensible Supervised LPP based on QR decomposition(ESLPP/QR). In the proposed algorithm, a dimension reduction algorithm of supervised locality preserving projection based on QR decomposition of training data matrix, namely SLPP/QR, is developed. It is efficient to solve the under-sampled problem. Using the discriminating information, the obtained SLPP/QR is combined with Fisher linear discriminant to receive final projection matrix and improve discriminant performance. Experimental results show that the algorithm has better discriminant performance than Principal Component Analysis(PCA) and LPP.

Key words: Principal Component Analysis(PCA), Locality Preserving Projection(LPP), QR decomposition, Fisher linear discriminant

摘要：

针对局部保留映射(LPP)算法不能提供数据集的差异信息问题，提出一种基于QR分解的扩展有监督LPP算法。该方法对训练数据矩阵进行QR分解，采用有监督的LPP算法进行降维，利用类别信息对降维后的数据进行Fisher线性判别式分析，得到最终的映射矩阵以提高判别性能。实验结果表明，该方法较主成分分析法和LPP方法有更好的判别性能。

关键词: 主成分分析, 局部保留映射, QR分解, Fisher线性判别式

CLC Number: 

• TP18