Piecewise Regularized Canonical Correlation Discrimination for Low-Resolution Face Recognition

Practical face recognition systems are sometimes confronted with low-resolution face images. Traditional super-resolution (SR) methods usually have limited performance because the target of SR may not consistent with that of classification, and time-consuming sophisticated SR algorithms are not suitable for real-time applications. We propose a piecewise regularized canonical correlation discrimination(rCCD) approach for LR face recognition without any SR preprocessing. The new method aims to maximize the canonical correlation between neighbor samples with different modes (i.e., low-resolution image and its high-resolution counterpart) while minimize the correlation between faraway modes. The experiments on publicly available databases show that our rCCD method indeed improves the recognition performance.