Multi-structure feature fusion for face recognition based on multi-resolution exaction
Xiaoli Ruan, Shunfang Wang, Shenshen Liu · 2016
In order to reduce the influence of illumination, pose and other external factors on the face recognition, a new multi-structure feature fusion method based on different resolutions is proposed in this paper. This method can extract facial features under different resolutions and fuse facial features in various structures to enhance the information content of the features. Specifically, first, the image is decomposed into three kinds of resolutions with different scales to form the multi-resolution image sequences, with Gabor wavelets used to extract the facial features of each image. Second, PCA is applied to reduce the dimension of the obtained high dimensional features. Third, every two or three features of the multi-resolution images are fused in various structures after they are extracted in three different scales respectively. Finally, linear discriminant analysis (LDA) method is utilized to train one of the most discriminative classifier after the features of dimensionality reduction are obtained. The experimental results based on the ORL facial database and YALE facial database showed that the proposed algorithm is highly effective and robust to treat the illumination and facial expression.