计算机工程与应用2019，Vol.55Issue(22)：140-145,6.DOI:10.3778/j.issn.1002-8331.1901-0338
基于余弦距离损失与卷积神经网络的人脸识别
Face Recognition Based on Cosine Distance Loss and Convolutional Neural Network
贾海波 1孟亮1
作者信息
- 1. 太原理工大学 信息与计算机学院,山西 晋中 030600
- 折叠
摘要
Abstract
Due to the recent development of convolutional neural networks in computer vision tasks, the performance of deep face recognition methods has been significantly improved. At present, the existing deep face models regard the face recognition task as a classification or metric learning task, aiming to learn the discriminative face features, but rarely achieve the characteristics of small intra-class distance and large inter-class distance. As a supervised signal, the loss func-tion plays an important role in learning face features in convolutional neural networks. In this paper, a Cosine Softmax Loss(CSL)based on the cosine distance is proposed, which makes the face feature more discriminative. In addition, experi-mental results on LFW and YTF using the same network architecture and training dataset show the superiority of the pro-posed method.关键词
人脸识别/卷积神经网络(CNN)/深度学习/余弦损失函数/度量学习Key words
face recognition/Convolutional Neural Network(CNN)/deep learning/cosine loss function/metric learning分类
信息技术与安全科学引用本文复制引用
贾海波,孟亮..基于余弦距离损失与卷积神经网络的人脸识别[J].计算机工程与应用,2019,55(22):140-145,6.
