摘要
基于L1范数的二维主成分分析是近年来提出的一种在图像域降维和特征提取的方法。通常,直接求解L1范数最大化问题很困难,因此,一种贪婪的策略被提出来了。然而,这种策略的初始化投影是随意选取的,为了获得更好的投影向量,得到一个最优的局部解,提出了一个非贪婪的L1范数最大化算法,该非贪婪算法具有三大优势:使用L1范数和非贪婪策略对于异常值很稳健;与PCA-L1相比较,更多的空间结构信息得到了保留;相比2DPCA-L1,所有的投影方向可以被优化并且可以获得更好的解决方案。人脸和其他数据集上的实验结果验证了所提出的方法更加有效。
2-D principal component analysis( 2-DPCA) based on L1-norm is a recently developed method for the robust dimensionality reduction and the feature extraction in the image domain. Normally,due to the difficulty of directly solving the L1-norm maximization problem,a greedy strategy is proposed. However,the initialization of the strategy is arbitrarily chosen. To obtain a better projection vector and an optimal local solution,a robust 2-DPCA maximization algorithm with non-greedy L1-norm is presented. The proposed algorithm has three major advantages: 1) it is robust to outliers for L1-norm and non-greedy strategy; 2) more spatial structure information is preserved compared with PCA-L1; 3) all projection directions can be simultaneously optimized and much better solution can be obtained than that of 2-DPCA-L1. Experimental results on face and other datasets confirm the effectiveness of the proposed algorithm.
出处
《南京邮电大学学报(自然科学版)》
北大核心
2016年第2期90-95,共6页
Journal of Nanjing University of Posts and Telecommunications:Natural Science Edition
基金
重庆市自然科学基金(CSTC2012jj A40054)
重庆市研究生科研创新(CYS14143)资助项目
关键词
二维主成分分析
L1范数
非贪婪算法
异常值
主成分分析法
2-D principal component analysis(2-DPCA)
L1-norm
non-greedy algorithm
outliers
princi pal component analysis(PCA)