Feature Extraction Based on Lp-norm Generalized Principal Component Analysis

Zhizheng Liang,Shixiong Xia,Yong Zhou,Lei Zhang,Youfu Li
DOI: https://doi.org/10.1016/j.patrec.2013.01.030
IF: 4.757
2013-01-01
Pattern Recognition Letters
Abstract:In this paper, we propose Lp-norm generalized principal component analysis (PCA) by maximizing a class of convex objective functions. The successive linearization technique is used to solve the proposed optimization model. It is interesting to note that the closed-form solution of the subproblem in the algorithm can be achieved at each iteration. Meanwhile, we theoretically prove the convergence of the proposed method under proper conditions. It is observed that sparse or non-sparse projection vectors can be obtained due to the applications of the Lp norm. In addition, one deflation scheme is also utilized to obtain many projection vectors. Finally, a series of experiments on face images and UCI data sets are carried out to demonstrate the effectiveness of the proposed method.
What problem does this paper attempt to address?