基于改进型加性余弦间隔损失函数的深度学习人脸识别

Deep Learning Face Recognition Based on Improved Additive Cosine Margin Softmax Loss Function

为了实现监控场景下的人脸识别,采集了监控视频中500个人每人2张人脸图像构成SVF(Surveillance Video Faces)测试集,包括500个正样本对,499000个负样本对。提出一种改进型加性余弦间隔损失函数,对加性余弦间隔损失函数进行改进,通过在特征与目标权重夹角的余弦值减去一个值,在特征与非目标权重夹角的余弦值加一个值,该值为0~1之间的数,通过实验选取最佳值,达到减小类内距离,拉大类间距离的目的。实验结果表明,与Softmax损失函数、乘性角度间隔损失函数及加性余弦间隔损失函数训练的人脸识别模型相比,该方法在监控场景测试集人脸识别准确率最高,为99.1%。

In order to realize face recognition in surveillance scenes,two faces of 500 people in the surveillance video were collected to form the Surveillance Video Faces(SVF)test set,including 500 positive sample pairs and 499000 negative sample pairs.An improved additive cosine margin softmax loss function is proposed to improve the additive margin softmax loss function.By subtracting a value from the cosine value of the angle between the feature and the target weight,and adding a value to the cosine value of the angle between the feature and the non-target weight,which can reduce the intra-class distance and enlarge the inter-class distance.The value is between 0 and 1,and the best value is selected by experiment.The experimental results show that compared with the face recognition model of the softmax loss function,the angular softmax loss function and the additive margin softmax loss function training,the method has the highest accuracy of face recognition in the monitoring scene test set,which is 99.1%.