Sparse eigenfaces analysis for recognition

Face recognition plays a fairly vital role in human computer inter-action. Eigenface algorithm is one representative approach that projects face images onto a low dimensional feature space using principal components analysis (PCA) to choose the maximal total scatter across all classes. However, due to the linear combination of original samples, eigenfaces may be difficult to interpret and explain. In this paper, we present sparse eigenfaces method for face recognition which employs sparse principal components analysis (SPCA) to find sparse eigenfaces capturing the maximal variance of all face images. Particularly, sparse eigenfaces can achieve reasonable trade-off between interpretability and the maximal variance projection of all face images. Finally, we carefully conduct face recognition experiments on Yale face data set. The experimental results demonstrate that sparse eigenfaces method outperforms the traditional eigenfaces method.