基于深度学习的动态人脸识别方法  

Dynamic face recognition with deep learning

作　　者：何元烈[1] 刘峰[1] 孙盛[1] HE Yuan-lie;LIU Feng;SUN Sheng(School of Computers,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学计算机学院

出　　处：《计算机工程与设计》2019年第11期3299-3305,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(41501362)

摘　　要：为更好地解决动态人脸识别在非受限环境下鲁棒性差的问题,提出基于深度学习的动态人脸识别方法。该方法结合迁移学习、多任务学习、增强学习和循环神经网络的优点,用预先训练好的模型提取视频每一帧的人脸图像特征,对其进行同时跟踪和识别,采用增强学习进一步提高识别效果,使用循环神经网络对一段视频进行识别。实验结果表明,该方法与当前先进方法相比,识别准确性接近最好成绩且鲁棒性有了较大提升。To better solve the problem of poor robustness of dynamic face recognition in unconstrained environment,a dynamic face recognition method based on deep learning was proposed.The advantages of transfer learning,multi-task learning,enhancing learning and recurrent neural network were combined in this method.The pre-trained model was used to extract the facial image features of each frame of the video,and they were simultaneously tracked and identified,recognition effects were further improved through enhanced learning.The recurrent neural network was used for a segment of video to implement face recognition.Experimental results show that compared with the state of the art methods,the recognition accuracy of the proposed method is close to the best performance and its robustness is greatly improved.

关 键 词：深度学习 动态人脸识别 迁移学习 多任务学习 增强学习 循环神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]