感知特征增强学习的低分辨率人脸识别
Low-resolution face recognition based on perceptual feature enhancement learning
投稿时间:2020-10-31  修订日期:2020-10-31
DOI:
中文关键词: 低分辨率人脸识别  深度学习  超分辨率重构  感知特征增强
英文关键词: Low resolution face recognition  Deep learning  Super-resolution reconstruction  Enhancement of perceptual features
基金项目:国家自然科学基金项目(面上项目,重点项目,重大项目)
作者单位E-mail
熊承义 中南民族大学 电子信息工程学院智能无线通信湖北省重点实验室 E-mail:xiongcy@mail.scuec.edu.cn 
邵奔 中南民族大学 电子信息工程学院智能无线通信湖北省重点实验室  
高志荣 中南民族大学 计算机科学学院  
柳霜 中南民族大学 电子信息工程学院智能无线通信湖北省重点实验室  
李雪静 中南民族大学 电子信息工程学院智能无线通信湖北省重点实验室  
摘要点击次数: 30
全文下载次数: 0
中文摘要:
      针对低分辨率人脸图像信息缺失而导致的人脸识别能力受限问题,提出了一种感知特征增强学习的低分辨率人脸识别深度网络.通过高分辨率图像特征引导低分辨率图像的超分辨率重建并监督识别网络的训练,有效改善了低分辨率人脸识别的性能.具体地,整个网络由两个通道组成.一个通道学习高分辨率图像特征分布,用于监督低分辨率人脸识别通道整个训练过程;另一通道实现对低分辨率图像的超分辨率重构,以及重构图像的特征提取与分类识别.身份损失与质量损失联合约束重构网络训练,有效的提高超分辨率网络和识别网络性能.实验结果验证了提出方法的可行性和有效性,特别在极低分辨率情况下也得到明显的性能提升.
英文摘要:
      A low resolution deep face recognition network based on perceptual feature enhancement learning is proposed to solve the problem of limited face recognition ability caused by the lack of low-resolution face image information. The performance of low-resolution face recognition is effectively improved by guiding the super-resolution reconstruction of low-resolution images and supervising the training of recognition network through high-resolution image features. Specifically, the whole network is composed of two channels. One channel learns the feature distribution of high-resolution images and is used to supervise the whole training process of low-resolution face recognition channels. The other channel realizes the super-resolution reconstruction of low-resolution images and feature extraction and classification recognition of reconstructed images. The combination of identity loss and quality loss can effectively improve the performance of super-resolution network and identification network. The experimental results show that the proposed method is feasible and effective, especially in the case of very low resolution, the performance is improved obviously.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭