改进的线性局部切空间排列算法

摘要

线性局部切空间排列算法(LLTSA)是一种能很好地适用于识别问题的非线性降维方法,但LLTSA仅仅关注了数据的局部几何结构,而没有体现数据的整体信息.提出了一种基于主成分分析(PCA)改进的线性局部切空间排列算法(P-LLTSA),该算法在LLTSA的基础上,考虑了样本的全局结构,进而得到更好的降维效果.在经典的三维流形和在MNIST图像库手写体识别的实验中,识别率较PCA、局部保持投影算法(LPP),LLTSA有明显提高,证实了该算法在识别问题中的有效性.%Linear Local Tangent Space Alignment (LLTSA) algorithm is a non-linear dimension reduction method which can be easily applied to recognition problems.It pays attention on the local geometric structure of data, but it neglects the global information of data.In this paper, an improved LLTSA algorithm based on principal component analysis (PCA) was proposed, and this method took the global structure of sample into consideration and contained a better reduction dimension result.In the classical experiment of 3D manifold and MNSIT image dataset script recognition, P-LLTSA has a higher recognition rate by contrast to PCA, LPP and LLTSA, which verifies the effectiveness of PLLTSA.