DCT和LBP特征融合的人脸识别

Fusing DCT and LBP Features for Face Recognition

提出一种特征融合的人脸识别新方法.该方法将人脸图像中少量的低频离散余弦变换(DCT)系数用作人脸的频域特征;把人脸图像规则地分成多个子块,计算每个子块的局部二值图(LBP)编码直方图.这些子块的LBP直方图连接成一个空域全局直方图,作为人脸的描述向量.这个描述向量经过PCA降维后作为人脸的LBP特征.DCT特征和LBP特征分别归一化,然后进行特征融合.在ORL人脸库上的实验显示了所提方法比单独采用DCT或LBP特征的人脸识别有较好的性能改善.

A novel method by fusing discrete cosine transform （DCT） and local binary pattern （LBP） features is proposed for face recognition in this research. The primary information of the face image was centralized in a small number of DCT coefficients, which were used as the frequency feature of the face. The face was divided regularly into small regions, from which LBP code histograms were computed and concatenated into a spatial global histogram used as descriptor vector of the face. The descriptor vector was dimensionally reduced by PCA. Then, the DCT features and the LBP features were fused posterior to the normalization. The experiments on ORL face database show the improvability of the proposed scheme on the methods using just single DCT or LBP feature for face recognition.