摘要
图像奇异值分解产生的归一化系数可以保留原始图像的大部分纹理信息并对光照条件不敏感,本文中采用归一化后的奇异值矩阵即奇异值脸来表示低光照人脸图像的主要特征.基于卷积神经网络的深度学习算法已经广泛应用于图像识别领域,通过在网络中加入注意力模块可以关注图像中的重要信息并抑制不必要的信息.本文提出一种结合奇异值脸和注意力卷积神经网络的人脸识别模型,算法首先采用归一化后的奇异值矩阵来表示人脸特征,然后将特征输入到添加了注意力模块的深度卷积神经网络中,通过跨通道和空间的信息融合提高网络的健壮性,最后通过网络的迭代训练完成人脸图像的分类识别.通过在两个常用数据库上的实验,证实了本文提出的算法具有更好的识别性能和更优的光照鲁棒性.
The normalization coefficients of singular value decomposition(SVD)can retain most of the texture information of the original image,and they are not sensitive to different lighting conditions.In this paper,the normalized singular value matrix is used to represent the main features of low light face image.The deep learning algorithm based on convolutional neural network has been widely used in the field of image recognition.By adding attention module in the network,we can pay attention to the important information in the image and suppress the unnecessary information.In this paper,a kind of novel face recognition algorithm is proposed,which combines singular value decomposition face(SVDF)and convolutional neural network with attention(ACNN).Firstly,the normalized singular value matrix is used to represent the face features.Secondly,the features of face images are input into the convolutional neural network with attention module,and the cross-channel and spatial information are fused to improve the robustness of the network.Finally,the input face images are classified and recognized.Experiments on two common databases show that our algorithm has better recognition performance and light robustness.
作者
朱娅妮
倪煊
姚晔
ZHU Ya-ni;NI Xuan;YAO Ye(College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310023,China;School of Cyberspace,Hangzhou Dianzi University,Hangzhou 310018,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2020年第8期1763-1767,共5页
Journal of Chinese Computer Systems
基金
浙江省科技厅重点研发计划项目(2017C01062)资助。
