基于卷积神经网络的带遮蔽人脸识别
|
摘要
基于卷积神经网络Inception-ResNet-v1模型进行训练与学习,实现了在添加遮挡干扰因素下的人脸识别。将图像嵌入到d维度的欧几里得空间,采用Triplet Loss作为损失函数,直接学习特征间的可分性。选取LFW(labeled faces in wild)数据集和摄像头采集的人脸图片制作训练集和测试集。结果表明,模型在眼部被遮挡的情况下识别率为98. 8%,在嘴部被遮挡的情况下识别率为98. 6%,在眼部和嘴部同时被遮挡的情况下识别率为96. 9%。模型在遮挡率为20%～30%时,识别率能够达到98. 2%。从实验结果可以得出,模型在一定遮挡的情况下能得到较好的识别效果。
In this paper,based on the convolutional neural network Inception-ResNet-v1 model,the training and learning were carried out to realize occluded face recognition. The model method was to embed the image into the Euclidean space of d dimension,and used Triplet Loss as the loss function to directly learn the separability among features. The experiments were conducted with the LFW(labeled faces in wild)data set and the face images collected by the camera. The results showed that the recognition rates of the model were 98. 8% in the case of eyes occlusion,98. 6% in the case of mouth occlusion and 96. 9% in the case of eyes and mouth both occlusion. The recognition rate was up to 98. 2% when the occlusion rate was 20%-30%. It can be concluded from the experimental results that the model can obtain better recognition results under certain occlusion conditions.
In this paper,based on the convolutional neural network Inception-ResNet-v1 model,the training and learning were carried out to realize occluded face recognition. The model method was to embed the image into the Euclidean space of d dimension,and used Triplet Loss as the loss function to directly learn the separability among features. The experiments were conducted with the LFW(labeled faces in wild)data set and the face images collected by the camera. The results showed that the recognition rates of the model were 98. 8% in the case of eyes occlusion,98. 6% in the case of mouth occlusion and 96. 9% in the case of eyes and mouth both occlusion. The recognition rate was up to 98. 2% when the occlusion rate was 20%-30%. It can be concluded from the experimental results that the model can obtain better recognition results under certain occlusion conditions.
引文
[1]XU W K,LEE S H,LEE E J.A robust method for partially occluded face recognition[J].KSII Transactions on Internet and Information Systems,2015,9(7):2667-2682.
[2]邓道举,贾楠,李秀梅.基于稀疏表达分类的人脸识别及算法对比研究[J].杭州师范大学学报(自然科学版),2018,17(3):319-325.
[3]徐书燕.基于稀疏表示和线性回归的人脸识别研究[D].广州:华南理工大学,2017.
[4]SHANMUGASUNDARAM K,SHARMA S,RAMASAMY S K.Face recognition with CLNF for uncontrolled occlusion faces[C]//Proceedings of 2016 IEEE International Conference on Recent Trends in Electronics,Information&Communication Technology,2016.
[5]SZEGEDY C,IOFFE S,VANHOUCKE V.Inception-v4,Inception-ResNet and the impact of residual connections on learning[C]//Proceedings of 31st AAAI Conference on Artificial Intelligence,2017:4278-4284.
[6]SCHROFF F,KALENICHENKO D,PHILBIN J.FaceNet:a unified embedding for face recognition and clustering[C]//Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition,2015.
[7]LECUN Y,BOTTOU L,BENGIO Y,et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.
[8]SZEGEDY C,IOFFE S,VANHOUCKE V,et al.Inception-v4,Inception-ResNet and the impact of residual connections on learning[C]//Proceedings of ICLR 2016 Workshop,2016.
[9]常亮,邓小明,周明全,等.图像理解中的卷积神经网络[J].自动化学报,2016,42(9):1300-1312.
[10]王晓刚.图像识别中的深度学习[J].中国计算机学会通讯,2015,11(8):15-23.
[11]卢宏涛,张秦川.深度卷积神经网络在计算机视觉中的应用研究综述[J].数据采集与处理,2016,31(1):1-17.
[12]冉鹏,王灵,李昕,等.改进SoftMax分类器的深度卷积神经网络及其在人脸识别中的应用[J].上海大学学报(自然科学版),2018,24(3):352-366.
[13]赵显达,黄欢.基于卷积神经网络的人脸识别的研究[J].信息技术,2018(9):15-23.
[14]刘帆,刘鹏远,李兵,等.TensorFlow平台下的视频目标跟踪深度学习模型设计[J].激光与光电子学进展,2017,54(9):283-291.
[15]李倩玉,蒋建国,齐美彬.基于改进深层网络的人脸识别算法[J].电子学报,2017,45(3):619-625.
[2]邓道举,贾楠,李秀梅.基于稀疏表达分类的人脸识别及算法对比研究[J].杭州师范大学学报(自然科学版),2018,17(3):319-325.
[3]徐书燕.基于稀疏表示和线性回归的人脸识别研究[D].广州:华南理工大学,2017.
[4]SHANMUGASUNDARAM K,SHARMA S,RAMASAMY S K.Face recognition with CLNF for uncontrolled occlusion faces[C]//Proceedings of 2016 IEEE International Conference on Recent Trends in Electronics,Information&Communication Technology,2016.
[5]SZEGEDY C,IOFFE S,VANHOUCKE V.Inception-v4,Inception-ResNet and the impact of residual connections on learning[C]//Proceedings of 31st AAAI Conference on Artificial Intelligence,2017:4278-4284.
[6]SCHROFF F,KALENICHENKO D,PHILBIN J.FaceNet:a unified embedding for face recognition and clustering[C]//Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition,2015.
[7]LECUN Y,BOTTOU L,BENGIO Y,et al.Gradient-based learning applied to document recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.
[8]SZEGEDY C,IOFFE S,VANHOUCKE V,et al.Inception-v4,Inception-ResNet and the impact of residual connections on learning[C]//Proceedings of ICLR 2016 Workshop,2016.
[9]常亮,邓小明,周明全,等.图像理解中的卷积神经网络[J].自动化学报,2016,42(9):1300-1312.
[10]王晓刚.图像识别中的深度学习[J].中国计算机学会通讯,2015,11(8):15-23.
[11]卢宏涛,张秦川.深度卷积神经网络在计算机视觉中的应用研究综述[J].数据采集与处理,2016,31(1):1-17.
[12]冉鹏,王灵,李昕,等.改进SoftMax分类器的深度卷积神经网络及其在人脸识别中的应用[J].上海大学学报(自然科学版),2018,24(3):352-366.
[13]赵显达,黄欢.基于卷积神经网络的人脸识别的研究[J].信息技术,2018(9):15-23.
[14]刘帆,刘鹏远,李兵,等.TensorFlow平台下的视频目标跟踪深度学习模型设计[J].激光与光电子学进展,2017,54(9):283-291.
[15]李倩玉,蒋建国,齐美彬.基于改进深层网络的人脸识别算法[J].电子学报,2017,45(3):619-625.