基于GLC-KSVD的稀疏表示人脸识别算法  

Sparse Representation Face Recognition Algorithm Based on Gabor Feature and Label Consistent K-SVD

作　　者：封睿[1] 李小霞[1] 

机构地区：[1]西南科技大学信息工程学院,四川绵阳621010

出　　处：《四川兵工学报》2014年第4期88-92,共5页Journal of Sichuan Ordnance

基　　金：西南科技大学博士基金(053109);四川省教育厅基金项目(11ZB106)

摘　　要：稀疏编码中的字典学习是基于稀疏表示图像分类的核心内容,为此提出了一种基于Gabor特征和标签一致K-SVD(GLC-KSVD)字典学习的稀疏表示人脸识别算法;由于Gabor特征对光照、表情和姿态等具有一定的鲁棒性,首先对图像进行Gabor特征提取,用增广的Gabor特征矩阵来构建初始字典,然后通过字典学习得到原子与类别标签相对应的判别性字典和线性分类器,字典学习模型综合了重建误差、分类误差和稀疏编码误差,通过字典的标签一致约束,同一类别的样本得到相似的编码系数;实验结果表明:该算法具有良好的识别精度和较高的识别效率。Dictionary learuing in sparse coding is an important content on image classification based on sparse representation. Therefore, a sparse representation face recognition algorithm was proposed based on Gabor feature and Label Consistent K-SVD （GLC-KSVD） dictionary learning. Considering that Gabor fea- ture is robust to variations of illumination, expression and pose, at first the image Gabor features were ex- tracted and used as the augmented Gabor feature matrix to construct the initial dictionary. Then an discrim- inative dictionary and linear classifier was learned simultaneously by the dictionary learning model, which combined the sparse-code error with the reconstruction error and the classification error, and the learned dictionary atoms were corresponded to the class labels. The feature points with the same class labels have similar sparse codes according to the label consistent constraint. Experiment results show that the proposed method has good recognition accuracy and higher recognition efficiency.

关 键 词：稀疏表示 人脸识别 GABOR特征 GLC-KSVD字典学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]