面向疫情防控的口罩重度遮挡人脸特征提取与识别研究  

Research on face feature extraction and recognition heavily occluded by masks for epidemic prevention and control

作　　者：栗科峰 熊欣 夏冰[2] LI Kefeng;XIONG Xin;XIA Bing(School of Electrical Information Engineering,Henan University of Engineering;School of Information Engineering,Zhengzhou Shengda University,Zhengzhou 451191,China)

机构地区：[1]河南工程学院电气信息工程学院 [2]郑州升达经贸管理学院信息工程学院,河南郑州451191

出　　处：《长江信息通信》2022年第9期13-15,共3页Changjiang Information & Communications

基　　金：河南省科技攻关项目(222102210268);河南省高等学校重点科研项目(22B51003)。

摘　　要：为解决疫情常态防控下人脸被口罩重度遮挡的识别难题,针对人脸关键特征点被口罩遮挡带来的识别精度下降问题,提出了一种改进的全局均匀二值模式-GUBP,引入相邻像素间的相关性来表征图像的细节纹理特征,通过全局均匀模式直方图提取人脸图像的特征向量,大大降低了由于面部重度遮挡带来的关键特征丢失的影响;使用一种新的组合距离分类方法-CDIS,综合相关距离、欧氏距离和马氏距离的优越性,通过计算三种距离的平方和的平方根,使口罩遮挡样本的分类误差最小化,获得了更加稳健的人脸分类度量标准;在MIT-CBCL人脸数据库添加不同比例口罩遮挡后进行的对比实验表明,即使口罩遮挡比例增加到50%,该方法依然可以获得87.15%的平均识别率,具有较强的鲁棒性和较高的识别效率。In order to solve the recognition problem that the face is heavily occluded by the mask under normal epidemic prevention,an improved global uniform binary pattern-GUBP is proposed to solve the problem of the recognition accuracy decline caused by the occlusion of the key feature points of the face by the mask. The correlation between adjacent pixels is introduced to describe the detailed texture features of the image, and the feature vector of the face image is extracted through the global uniform pattern histogram, which greatly reduces the impact of the loss of key features due to heavy facial occlusion. Using a new combined distance classification method-CDIS, which combines the superiority of Correlation distance, Euclidean distance and Mahalanobis distance. The classification error of masks occluded samples is minimized by computing the square root of the sum of the squares of the three distances. This method obtains more robust face classification criteria. After adding different proportions of mask occlusions to the MIT-CBCL face database,comparative experiments show that even if the mask occlusion ratio increases to 50%, the method can still obtain an average recognition rate of 87.15%. The algorithm in this paper has strong robustness and high recognition efficiency.

关 键 词：口罩重度遮挡 全局均匀二值模式 组合距离 MIT-CBCL人脸数据库 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]