2010 International Conference on Pattern Recognition
Sparse Local Discriminant Projections for Face Feature Extraction Zhihui Lai1, Zhong Jin1, Jian Yang1, W.K Wong2 1
School of Computer Science, Nanjing University of Science and Technology, Nanjing, Jiangsu, P. R. China, 210094 2 Institute of Textiles & Clothing,The Hong Kong Polytechnic University, Hong Kong E-mail:
[email protected]; zhongjin @mail.njust.edu.cn;
[email protected];
[email protected]
Abstract—One of the major disadvantages of the linear dimensionality reduction algorithms, such as Principle Component Analysis (PCA) and Linear Discriminant Analysis (LDA), are that the projections are linear combination of all the original features or variables and all weights in the linear combination known as loadings are typically non-zero. Thus, they lack physical interpretation in many applications. In this paper, we propose a novel supervised learning method called Sparse Local Discriminant Projections (SLDP) for linear dimensionality reduction. SLDP introduces a sparse constraint into the objective function and obtains a set of sparse projective axes with directly physical interpretation. The sparse projections can be efficiently computed by the Elastic Net combining with spectral analysis. The experimental results show that SLDP give the explicit interpretation on its projections and achieves competitive performance compared with some dimensionality reduction techniques. Keywordsf-feature extraction; projections; physical interpretation;
Elastic
characterize the distribution of the data and the label information is not made good use. As a result, the discriminant ability of USSL is limited. Motivated by the manifold learning and sparse subspace learning algorithms mentioned above, in this paper, we propose a novel sparse subspace learning framework called Sparse Local Discriminant Projections (SLDP) to overcome the drawback of USSL. SLDP aims to preserve the local within-class neighborhood relationships of the data set and simultaneously maximizes the interclass separability with sparse constraint. In SLDP, in order to improve the classification accuracy, two kinds of different characterizations are combined together to reflect the data distribution. The sparse projections can be efficiently computed by the Elastic Net. The rest of the paper is organized as follows. SLDP algorithm is described in Section 2. In Section 3, experiments are carried out to evaluate our SLDP algorithm. Finally, the conclusions and future work are given in Section 4.
Net ; sparse
II. I.
INTRODUCTION
Let us consider a set of m samples {x1 , x2 ,..., xm } taking values in an n -dimensional space, and assume that each sample belongs to one of c classes. Let us also consider a linear transformation mapping from the original n -dimensional space into an d -dimensional feature space, where d