Scale-robust feature extraction for face recognition

In video surveillance, the sizes of face images are very small. However, few works have been done to investigate scale-robust face recognition. Our experiments on appearance-based methods in different resolutions show that such methods as Neighboring Preserving Embedding (NPE) and Locality Preserving Projections (LPP) preserving local structure of data are less effective than the methods retaining global structure, for example, Linear Discriminant Analysis (LDA) and Principal Component Analysis (PCA) under low-resolution condition. Based on these underlying phenomena, we propose a new graph embedding method named FisherNPE holding both global and local structures of data for scale-robust feature extraction. Experimental results on ORL and Yale database indicate that our method obtains good results on both low- and high-resolution images.