摘要
针对常规深度学习的人脸识别方法训练数据量巨大和训练难收敛的问题,提出一种基于残差网络与中心损失的人脸识别方法。利用生成对抗网络方法解决训练数据分布不均衡问题,利用数据增强方法解决数据不足问题;改进残差网络,使其匹配较小数据集,解决训练难收敛问题;将交叉熵损失与中心损失结合,作为模型训练过程中的监督信号,使类间分散、类内聚合。实验结果表明,在小数据的前提下,识别算法能够准确地识别出人脸。模型在测试集上的准确率达97.46%。
Existing face recognition algorithms based on deep learning algorithms are difficult to train because they require huge amount of training dataset.A method based on Resnet and center loss was proposed.The unbalanced distribution of data was resolved using generative adversarial nets and the insufficiency of data was solved by data augmentation.Resnet was improved for small dataset and the difficult problem of training convergence was resolved.The cross-entropy loss and center loss were combined as the supervisory signal in the model training that contributed to inter-class dispersion and intra-class aggregation.With the limit of small dataset,experimental results show that the testing-accuracy rate achieves 97.46%.
引文
[1]WAN Yuan,LI Huanhuan,WU Kefeng,et al.Fusion with layered features of LBP and HOG for face recognition[J].Journal of Computer-Aided Design&Computer Graphics,2015,27(4):640-650(in Chinese).[万源,李欢欢,吴克风,等.LBP和HOG的分层特征融合的人脸识别[J].计算机辅助设计与图形学学报,2015,27(4):640-650.]
[2]Xian X B,Wu H J,Zhang M X,et al.A novel multi-pose face recognition via robust sift feature[C]//International Conference on Wavelet Analysis and Pattern Recognition.IEEE,2013:32-37.
[3]WU Lifang,HOU Yaxi,XU Xiao,et al.Compact complete pose binary SIFT for face recognition with pose variation[J].Chinses Journal of Scientific Instrument,2015,36(4):736-742(in Chinese).[毋立芳,侯亚希,许晓,等.基于紧致全姿态二值SIFT的人脸识别[J].仪器仪表学报,2015,36(4):736-742.]
[4]LIANG Shufen,LIU Yinhua,LI Lishen.Face recognition algorithm based on LBP and deep learning under unconstrained conditions[J].Journal of Communications,2017,35(6):154-160(in Chinese).[梁淑芬,刘银华,李立琛.基于LBP和深度学习的非限制条件下人脸识别算法[J].通信学报,2017,35(6):154-160.]
[5]Parkhi OM,Vedaldi A,Zisserman A.Deep face recognition[C]//InProc.BMVC,2015.
[6]Wen Y,Zhang K,Li Z,et al.A discriminative feature learning approach for deep face recognition[C]//European Conference on Computer Vision.Cham:Springer,2016:499-515.
[7]Schroff F,Kalenichenko D,Philbin J.Facenet:A unified embedding for face recognition and clustering[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,2015:815-823.
[8]He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,2016:770-778.
[9]Nech A,Kemelmacher-Shlizerman I.Level playing field for million scale face recognition[C]//IEEE Conference on Computer Vision and Pattern Recognition.IEEE,2017:3406-3415.
[10]Ioffe S,Szegedy C.Batch normalization:Accelerating deep network training by reducing internal covariate shift[C]//International Conference on International Conference on Machine Learning.JMLR.org,2015:448-456.
[11]Srivastava N,Hinton G,Krizhevsky A,et al.Dropout:Asimple way to prevent neural networks from overfitting[J].The Journal of Machine Learning Research,2014,15(1):1929-1958.
[12]Szegedy C,Zaremba W,Sutskever I,et al.Intriguing properties of neural networks[J].Computer Science,2013:1-10.
[13]Goodfellow I J,Pouget-Abadie J,Mirza M,et al.Generative adversarial nets[C]//International Conference on Neural Information Processing Systems.MIT Press,2014:2672-2680.
[14]Li Y,Xiao N,Ouyang W.Improved boundary equilibrium generative adversarial networks[J].IEEE Access,2018(6):11342-11348.