摘要
针对采用融合深度哈希的卷积神经网络进行人脸识别时可能存在准确率下降及内存占用率偏高的问题,提出了基于权重哈希化的深度人脸识别算法。首先,提出一种基于高低维特征维度拼接的全卷积深度哈希网络,用以保证融合深度哈希后网络模型的识别准确率;然后,提出一种基于权重哈希化的模型压缩方法,将浮点型权重量化为哈希编码来进行模型存储,用以减少模型的内存占用率。实验表明,该方法在基于VGG框架进行改进时,可将VGG原网络的识别总效率提高68%,将准确率提高1.67%且使模型尺寸压缩了91.2%;该方法扩展到Sphereface框架时,在准确率略有提升的情况下将识别效率提高了61%,将模型压缩了42.24%。因此所提方法可提高识别准确率和效率,并减少内存占用率,同时还可扩展应用于其他网络。
In order to solve the problem that the accuracy rate may decrease and the memory occupancy rate may still be high when the convolution neural network with fused depth hash is used for face recognition,this paper proposed a deep face recognition algorithm based on weighted hashing.Firstly,a fully convolutional neural network of deep hash based on dimension splicing with high and low dimensional features is proposed to improve recognition accuracy.Secondly,a model compression method with floating-point weights quantized into hash coding is proposed to reduce memory occupancy rate of the model.The experimental results show that the proposed method improves efficiency by 68%,improves the Rank-1 accuracy by 1.67%,and the model size is compressed by 91.2% when it is improved based on VGG framework.In addition,it improves efficiency by 61% when the Rank-1 accuracy is slightly improved,and the model size is reduced by 42.24% when it is improved based on Sphereface framework.The results indicate that the proposed method can improve the recognition accuracy and efficiency,and reduce the memory usage.It also can be applied for other frameworks.
引文
[1] TAIGMAN Y,YANG M,RANZATO M,et al.DeepFace:Clo- sing the Gap to Human-Level Performance in Face Verification[C]//IEEE Conference on Computer Vision and Pattern Recognition.Columbus:IEEE Computer Society,2014:1701-1708.
[2] TAIGMAN Y,YANG M,RANZATO M,et al.Web-scale trai- ning for face identification[C]//Proceedings of the IEEE Confe-rence on Computer Vision and Pattern Recognition.Boston:IEEE,2015:2746-2754.
[3] PARKHI O M,VEDALDI A,ZISSERMAN A.Deep Face Re- cognition[C]//British Machine Vision Conference.Swansea:BMVC,2015:41.1-41.12.
[4] WEN Y,ZHANG K,LI Z,et al.A Discriminative Feature Learning Approach for Deep Face Recognition[M]//Computer Vision-ECCV 2016.Amsterdam:Springer International Publi-shing,2016:499-515.
[5] LIU W,WEN Y,YU Z,et al.SphereFace:Deep Hypersphere Embedding for Face Recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition.Hawaii:IEEE Computer Society,2017:6738-6746.
[6] DENG J,GUO J,ZAFEIRIOU S.ArcFace:Additive Angular Margin Loss for Deep Face Recognition[OL].http://arxiv.org/abs/1801.07698.
[7] HE K,ZHANG X,REN S,et al.Deep Residual Learning for Image Recognition[C]//Computer Vision and Pattern Recognition.Las Vegas:IEEE,2016:770-778.
[8] 曾燕,陈岳林,蔡晓东.结合全局与局部池化的深度哈希人脸识别算法[OL].http://www.cnki.com.cn/Article/CJFDTotal-XDKD20180320000.htm.
[9] LI F,ZHANG B,LIU B.Ternary Weight Networks[C]//IEEE Conference on Computer Vision and Pattern Recognition.2016.
[10] YI D,LEI Z,LIAO S,et al.Learning Face Representation from Scratch[C]//IEEE Conference on Computer Vision and Pattern Recognition.Columbus:IEEE Computer Society,2014.
[11] HUANG G B,RAMESH M,BERG T,et al.Labeled faces in the wild:A database for studying face recognition in unconstrained environments:07-49 [R].Amherst:Technical Report,2007.
[12] BEAT-ROWDEN L,HAN H,KLARE B F,et al.Unconstrained Face Recognition:Identifying a Person of Interest From a Media Collection[J].IEEE Transactions on Information Forensics & Security,2014,9(12):2144-2157.
[13] ZHANG K,ZHANG Z,LI Z,et al.Joint Face Detection and Alignment Using Multi-task Cascaded Convolutional Networks[J].IEEE Signal Processing Letters,2016,23(10):1499-1503.