基于深度残差网络的光照干扰人脸识别方法  

A Facial Recognition Method Based on Deep Residual Network against Lighting Interference

作　　者：刘静[1] LIU Jing(Anhui International Studies University,Hefei 230012,China)

机构地区：[1]安徽外国语学院,安徽合肥230012

出　　处：《通化师范学院学报》2024年第10期42-47,共6页Journal of Tonghua Normal University

摘　　要：自然光照条件下的变化非常复杂,包括光照强度、方向、色温,以及阴影等因素的变化,这些因素都可能对人脸图像的外观产生显著影响.为此,提出了一种基于深度残差网络的光照干扰人脸识别方法.该方法引入残差学习框架,利用深度残差网络重建光照干扰下的人脸图像.通过应用优化白平衡技术,实现对人脸的光照不均衡的补偿.此外,改进后的卷积神经网络可以在深度学习框架中实现人脸识别,用于识别和验证个体身份.实验结果表明:该研究方法能够对人脸实现重建,且对处理光照干扰人脸识别的效果表现理想,损失值较低.The changes under natural lighting conditions are very complex,including changes in lighting intensity,direction,color temperature,and shadows,which may have a significant impact on the appearance of facial images.To this end,a facial recognition method based on deep residual networks against lighting interference is proposed.A residual learning framework is introducecd,utilizing deep residual networks,to reconstruct facial images under illumination interference.Optimized white balance technology is applied to compensate for uneven lighting on the face.Convolutional neural networks are improved to achieve facial recognition,and deep learning frameworks are used to identify and verify individual identities.The experimental results show that the research method can achieve facial reconstruction and has an ideal effect on facial recognition against lighting interference,with lower loss values.

关 键 词：深度残差网络 光照干扰 人脸识别 改进卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]