Schatten-p Norm Based Linear Regression Discriminant Analysis for Face Recognition

Locality-regularized linear regression classification (LLRC) shows good performance on face recognition. However, it sorely performs on the original space, which results in degraded classification efficiency. To solve this problem, we propose a dimensionality reduction algorithm named schatten-p norm based linear regression discriminant analysis (SPLRDA) for image feature extraction. First, it defines intra-class and inter-class scatters based on schatten-p norm, which improves the capability to deal with illumination changes. Then the objective function which incorporates discriminant analysis is derived from the minimization of intra-class compactness and the maximization of inter-class separability. Experiments carried on some typical databases validate the effectiveness and robustness of our method.