基于局部二值模式和级联AdaBoost的多模态人脸识别

Multi-modal face recognition based on LBP and chain AdaBoost

提出了一种基于局部二值模式(LBP)和级联AdaBoost的多模态人脸识别方法。采用级联AdaBoost算法分别从人脸深度图像和灰度图像的大量区域LBP直方图(RLBPH)中选出最有利于分类的少量特征,并连接成一个直方图向量,再分别用线性判别分析构建相应的线性子空间,用余弦相似度作为投影向量的相似度量,用求和规则进行信息融合。在FRGC数据库上的实验结果表明,提出的方法采用少量的特征取得了很好的识别效果,等错误率仅为1.40%。

A method combining Local Binary. Pattern （LBP） descriptor with chain AdaBoost was presented for muhimodal face recognition. Thousands of Regional LBP histograms （RLBPH） were generated from grey-level and depth face images respectively. Chain AdaBoost was utilized to select most informative RLBPHs. The selected RLBPHs were concatenated to a whole histogram. Then the corresponding linear subspaces were constructed by Linear Discriminant Analysis （LDA） respectively. The eosine similarity was adopted as the similarity metric of projected vectors. Sum rule was used to fuse 2D and 3D information. The experimental results on FRGC database demonstrate that the proposed method achieves high recognition performance with very. few features. The Equal Error Rate （EER） is only 1.40%.