Gabor-Feature Hallucination based on Generalized Canonical Correlation Analysis for face recognition

In face recognition, image resolution is an important factor which has a great influence on the recognition rate. In traditional face recognition for low-resolution images, face interpolation/super-resolution is usually performed first, and the constructed high-resolution face image will then pass through a face recognition system, which includes feature extraction and classification. To achieve a more efficient and accurate approach, we propose a new method of “Gabor-Feature Hallucination”, which predicts the high-resolution Gabor features from the low-resolution Gabor features directly by using linear regression and Generalized Canonical Correlation Analysis. Then, the low-resolution features in the projected Generalized Canonical Correlation space and the predicted high-resolution Gabor features are adopted for face classification. Our algorithm can therefore avoid performing interpolation/super-resolution and high-resolution Gabor feature extraction. Experimental results show that the proposed method has a superior recognition rate and efficiency to the traditional methods.