基于Zernike特征提取和LDA的人脸图像识别方法  

Face Recognition Method Based on Zernike Feature Extraction and LDA

作　　者：华珊珊[1] 

机构地区：[1]合肥学院计算机科学与技术系

出　　处：《控制工程》2018年第2期198-203,共6页Control Engineering of China

摘　　要：针对现有人脸识别方法不能很好的处理面部表情、光照、头部姿势和遮挡等问题,提出一种基于Zernike特征提取和线性判别分析（LDA）分类器的人脸识别方法。首先,对图像进行归一化处理,并进行裁剪;然后,利用Zernike矩（ZM）提取全局特征,并利用主成分分析（PCA）来对特征进行降维;最后,利用提出的决策融合算法对局部和全局特征进行融合,获得一个最终的特征向量作为LDA分类器的输入,进而进行人脸识别。在ORL、AR和CASIA 3D人脸数据库上的实验结果表明,该方法在具有较低计算量的同时显著提高了的人脸识别精度。For the issues that the existing face recognition method can not deal with the problem of facial expression, illumination, head pose and occlusion, this paper proposes a face recognition method based on Zernike feature extraction and linear discriminant analysis （LDA）classifier. Firstly, this method normalizes and cuts the image. Then, it uses the Zernike moment（ZM） to extract the global feature, and the feature is reduced by principal component analysis （PCA）. Finally, it uses the decision fusion algorithm to fuse the local and global features, and then obtains a final feature vector as the input of the LDA classifier. Experimental results on ORL, AR and CASIA 3D face databases show that the proposed method has lower computation while significantly improves the accuracy of face recognition.

关 键 词：人脸识别 ZERNIKE矩 PCA特征降维 LDA分类器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]