融合子区域局部二值模式特征与深层聚合网络的人脸识别

针对深度网络对人脸噪声敏感,且学习过程容易忽视人脸结构信息的问题,提出融合子区域局部二值模式(local binary pattern,LBP)特征和深层聚合网络的人脸识别算法。将人脸图像划分为不同子区域,并采用局部二值模式对人脸进行预处理,获取子区域人脸的LBP特征。不同子区域LBP特征输入不同的稀疏自动编码器,实现深层特征提取;然后不同稀疏自动编码器的输出特征通过全连接方式实现特征聚合,获得人脸特征向量用于分类。通过大量实验获取了最优的聚合网络模型架构和网络参数取值,改善了人脸识别效果。

Face recognition based on fusion of sub-region Local Binary Pattern features and deep aggregation network

Aiming at the problem that the depth network is sensitive to human face noise and the learning process is easy to ignore the information of face structure, this paper proposes a face recognition algorithm based on LBP features and deep aggregation network. The face image is divided into different sub-regions, and the local binary image is used to pre-process the face image in order to obtain the LBP features of the sub-region face. LBP features of different sub-regions are input to different sparse auto encoders to achieve deep feature extraction. Then, the output features of different sparse auto encoders are fully aggregated to obtain the feature vectors for classification. Through a large number of experiments, we get the optimal aggregation network model structure and network parameter values, and improve the face recognition effect.