不同光照强度变化下融合深度学习的人脸识别技术研究  

Research on face recognition technology based on fusion depth learning under different light intensity changes

作　　者：杨彦青[1] 宋星[1] YANG Yanqing;SONG Xing(Taizhou Vocational and Technical College,Taizhou Zhejiang 318000,China)

机构地区：[1]台州职业技术学院,浙江台州318000

出　　处：《自动化与仪器仪表》2023年第4期49-52,57,共5页Automation & Instrumentation

基　　金：2022年浙江省教育厅一般科研项目《不同光照强度变化下融合深度学习的人脸识别技术研究》(Y202249931)。

摘　　要：人脸识别是一种生物识别技术,通过人脸特征信息提取实现身份验证,是计算机视觉领域的热门问题之一。但当前的人脸图像识别受不同光照强度变化的干扰较大,识别技术亟待优化。针对不同强度变化下的人脸图像,提出了一种光照强度恢复方法进行图像预处理,并将与新的四元组中心损失函数相融合应用于人脸图像的识别中。结果表明,在Yale和ORL人脸数据库中,该方法的准确率最高分别达到了95%和96%;在LFW数据集和YTF数据集中,准确率提高的最大值为17.05%;在所选的人脸样本中,其识别准确率稳定在94%以上,具有较高的准确率和稳定性,为进一步提高人脸识别效果提供技术参考。Face recognition is a biometric technology.It is one of the hot issues in the field of computer vision to achieve identity verification by extracting facial feature information.However,the current face image recognition is greatly disturbed by different light intensity changes,and the recognition technology needs to be optimized.A light intensity restoration method is proposed to preprocess the face images under different intensity changes,and the fusion with the new quad central loss function is applied to face image recognition.The results show that the accuracy of this method reaches 95%and 97%respectively in Yale and ORL face databases;In LFW dataset and YTF dataset,the maximum increase in accuracy is 17.05%;In the selected face samples,the recognition accuracy is stable at more than 94%,with high accuracy and stability,providing reliable technical support for the further development of face recognition technology.

关 键 词：深度学习 人脸识别 光照恢复 损失函数 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]