复杂光照条件下的多姿状鲁棒性人脸识别方法  

A robust face recognition method under complex illumination conditions

作　　者：刘海山[1] 陈前军[1] LIU Haishan;CHEN Qianjun(Hubei University,Wuhan 430062,China)

机构地区：[1]湖北大学,武汉430062

出　　处：《激光杂志》2021年第2期42-46,共5页Laser Journal

基　　金：湖北省教育厅科学技术计划项目(No.B20176501)。

摘　　要：为改善复杂光照条件下的多姿状鲁棒性人脸识别的效果,提出了小波变换与LBP的多姿状鲁棒性人脸识别方法。通过二维离散小波变换对人脸图像进行二级小波分解提取到低频特征信息分量,并以重构初始图像的方式实现降噪滤波处理,滤除低频光照分量后完成复杂光照补偿;继续分解复杂光照补偿后的图像,采用LBP算子对子图像的鲁棒性部分纹理特征进行描述后,提取出人脸图像各子图像的直方图特征并连接,得到人脸LBP纹理特征,通过统计法运算该特征距离,并通过K近邻分类器实现人脸特征分类识别。以Yale-B与AR人脸库为测试对象,结果表明,所研究方法对复杂光照鲁棒性较强,识别人脸的准确率与效率较高,整体识别效果较好。To improve the performance of face recognition under complex illumination conditions,a face recognition method combining wavelet transform and LBP operator is proposed.The 2-d discrete wavelet transform was used to extract the face image’s low-frequency feature information component by second-order wavelet decomposition,and the initial image was reconstructed to realise noise reduction filtering.After filtering the low-frequency light component,the complex light compensation was completed.Continue to break down complex illumination compensation after image;the use of LBP operator for robustness is part of the image texture features are described.After extracting the human face image histogram of each image feature and connected into from LBP texture feature,through statistical computing,the characteristic distance,and facial features are implemented by K neighbour classifier classification recognition.Yale-b and AR face database were used as test objects.The results showed that the proposed method was robust to complex illumination,had higher accuracy and efficiency in face recognition,and had better overall recognition.

关 键 词：复杂光照 多姿状人脸识别 小波变换 LBP算子 降噪滤波 光照补偿 

分 类 号：TN929[电子电信—通信与信息系统]