考虑光照鲁棒性的超分辨率人脸识别系统设计  

Design of super-resolution face recognition system considering the illumination robustness

作　　者：郭瞻[1] 肖祖铭[1] GUO Zhan;XIAO Zuming(Jingdezhen University,Jingdezhen Jiangxi 343000,China)

机构地区：[1]景德镇学院,江西景德镇343000

出　　处：《激光杂志》2023年第5期224-230,共7页Laser Journal

基　　金：江西省教育厅科学技术研究项目(No.GJJ212822)。

摘　　要：为保障不同光照下低分辨率人脸的超分辨率识别精度与效率,设计了考虑光照鲁棒性的超分辨率人脸识别系统。通过包含DSP单元与ARM单元的控制器模块,驱动人脸视频采集模块。采集不同复杂光照的人脸视频信息并解析成视频帧后,预估视频帧序列的位移情况,恢复视频帧序列的超分辨率,融合超分辨率视频帧构成人脸图像样本。利用人脸特征提取模块补偿全部人脸图像样本复杂光照,并提取其LBP特征构成人脸库。通过人脸识别模块匹配人脸图像的LBP特征与人脸库,识别出超分辨率人脸图像。结果表明,该系统的光照鲁棒性人脸图像采集与人脸图像LBP特征提取两部分的实现效果均较好。可有效识别出背光、强光及弱光下的超分辨率人脸,识别效率较高,识别成功率能够达到96.7%,为光照鲁棒性人脸识别提供保障。In order to ensure the super-resolution recognition accuracy and efficiency of low-resolution faces un-der different illuminations,a super-resolution face recognition system considering illumination robustness is designed.Through the controller module containing DSP unit and ARM unit,the face video acquisition module is driven.After collecting and analyzing the face video information of different complex illuminations into video frames,the displace-ment of video frame sequence is estimated,the super resolution of video frame sequence is restored,and the super res-olution video frame is fused to form face image sample.The face feature extraction module is used to compensate the complex illumination of all face image samples,and LBP features are extracted to form a face library.The face recog-nition module is used to match the LBP features of real-time face images with the face database,and the super-resolu-tion face images are recognized.The results show that the illumination robustness of the system and LBP feature extrac-tion of face image are both effective,and the super-resolution face under backlight,strong light and weak light can be effectively recognized with high recognition efficiency,and the recognition success rate can reach 96.7%,which pro-vides guarantee for the illumination robust face recognition.

关 键 词：光照鲁棒性 超分辨率 人脸识别 控制器 人脸库 特征提取 

分 类 号：TN391[电子电信—物理电子学]