基于距离限定优化的多姿态人脸图像智能识别  

Intelligent Recognition of Multi Pose Face Images Based on Distance Constrained Optimization

作　　者：杨小琴[1] 朱玉全[2] YANG Xiao-qin;ZHU Yu-quan(Pujiang Institute,Nanjing Tech University,Nanjing Jiangsu 211134,China;School of Computer Science and Communication Engineering,Jiangsu University,Zhenjiang Jiangsu 212013,China)

机构地区：[1]南京工业大学浦江学院,江苏南京211134 [2]江苏大学计算机科学与通信工程学院,江苏镇江212013

出　　处：《计算机仿真》2022年第1期200-203,282,共5页Computer Simulation

基　　金：江苏省现代教育技术研究课题(2019-R-81745);江苏高校哲学社会科学研究一般项目(2019SJA2068)。

摘　　要：针对人脸角度、姿态和表情复杂多变,导致人脸面部表情识别准确率较低的问题,提出基于距离限定优化的多姿态人脸图像智能识别方法。根据人脸图像特征类别进行标号并调整图像角度等值,借助向量内积确定人脸图像特征相似程度,提取人脸图像特征;采用离散化处理人脸图像特征,构建特征矩阵,将人脸图像特征分量中维度作为条件属性,采用粗糙集近似约简法对所得特征向量降维处理;通过边界值和阈值优化特征向量距离,采集不同姿态特征向量,生成多种姿态人脸图像库,将待测人脸图像特征与人脸特征姿态库中图像相匹配,通过匹配结果相似程度判定二者是否为同一对象,完成人脸识别全过程。实验结果证明,采用所提识别方法对多姿态人脸图像识别的准确率较高,鲁棒性更好。Aiming at the problem of low accuracy of facial expression recognition due to the complexity of face angle, pose and expression, a multi pose face image intelligent recognition method based on distance limited optimization is proposed in the paper. Based on the feature category of face image, the image angle equivalence was labeled and adjusted. The similarity of facial image features was determined by vector inner product for extracting facial image features. The face image features were discretized. The characteristic matrix was founded. The dimension of face image feature component was set as conditional attribute. Rough set approximate reduction method was introduced to reduce the dimension of the feature vector. According to the boundary value and threshold, the eigenvector distance was optimized. Different pose feature vectors were collected to generate a variety of pose face image databases. The features of the face image to be tested matched the images in the face feature pose database. According to the matching results, whether they were the same object was determined, completing the whole process of face recognition. The experimental results show that this method has high recognition accuracy and excellent robustness.

关 键 词：特征向量距离 粗糙集近似约简法 特征降维 特征匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]