融合CDI和LBP的人脸特征提取与识别算法

Face recognition using fusion of cartesian differential invariant and LBP

针对光照、姿态和表情对人脸识别率造成严重影响的问题,提出了结合笛卡儿微分不变量(CDI,cartesian differential invariant)和LBP(local binary patterns)的人脸特征抽取与识别算法。首先,利用高斯微分算子抽取人脸图像的微分结构,组合这些微分结构得到一个不可约简的笛卡儿CDI集。其次,对CDI集中每个分量分别计算其LBP特征,并将所有分量的LBP特征连接起来以得到人脸图像的特征。最后,运用所抽取出的人脸局部描述特征和支持向量机(SVM)分类器完成人脸图像分类与识别。试验分析表明,基于CDI的LBP特征对人脸位置、姿态、光照和表情的变化具有较高的不变性。该算法在ORL和Yale人脸库中分别取得了98.5%和98.89%的识别率。

The performance of face recognition is seriously affected by illumination,pose,and expression.To solve this problem,a method for face features extraction and recognition is proposed based on the fusion of cartesian differential invariant（CDI） and LBP.Firstly,an irreducible Cartesian differential invariant set is obtained by combining the differential structure of the face images acquired by using Gauss differential operator.Secondly,by computing and linking all the LBP features of every item in the invariant set,the features of a face image are extracted.Finally,the classification and recognition facial image is completed by using （SVM） classifier and the extracted features.Experiments show that the LBP features based on Cartesian differential invariant has the merit of preferable invariant to face position,pose,illumination and expression variations.The method achieveds 98.5% recognition rate on ORL face database and 98.89% on Yale face database.