A Step-wise Refinement Algorithm for Face Recognition Based on Blocking Wavelet Transforms.
Yibing Wang, Zhou Tian-yun, Bang-Jun Hu · J. Inf. Hiding Multim. Signal Process. · 2015
In this paper, a new human face recognition algorithm is proposed. Blocking wavelet transforms are used for local features extraction, and serial integration of classi- fiers is adopted for final verification. The proposed technique consists of three stages. 1) A simple and efficient method is presented to extract global features, so as to obtain the mean face image. 2) In order to fully extract local features and overcome the problem of the small sample size, the concept of blocking wavelet transforms is presented. First of all, the image is divided into several non-overlapping blocks, and wavelet coefficients are obtained by using wavelet transform for each sub-block. Then, an improved dimen- sionality reduction technique, called Bidirectional two Dimensional Principal Component Analysis (B2DPCA) is used to reduce dimensionality. 3) To further improve the recog- nition performance, the global and local facial features are combined in a serial manner. Global features are used for coarse classification, and the global and local features are integrated for fine classification. Comprehensive experiments on three data sets (ORL, AR and FERET) demonstrate the effectiveness of our scheme.