Low-Frequency Representation for Face Recognition

This paper proposes a low-frequency representation (LFR) method for face images based on support value transform. LFR works directly on 2D image matrices rather than 1D vectors, thus the image matrix does not need to be transformed into a vector prior to feature extraction. In LFR, the simple and slowly variational features for face images are remained. To demonstrate the effectiveness of LFR, a series of experiments are performed on two face image databases: ORL and UMIST face databases. Experimental results indicate that LFR provides a better representation for face images with multi-view and slightly various illumination.