基于LVQ神经网络的人脸朝向识别算法  

Face Orientation Recognition Based on LVQ Neural Network

作　　者：焦涌 JIAO Yong(The 28th Research Institute of China Electronics Technology Group Corporation,Nanjing 210023,China)

机构地区：[1]中国电子科技集团公司第二十八研究所,南京210023

出　　处：《指挥信息系统与技术》2022年第4期76-79,84,共5页Command Information System and Technology

基　　金：装备发展部“十三五”预研课题(315055101)资助项目。

摘　　要：为了解决业界流行的人脸朝向识别算法人脸识别准确率较低的问题,提出了一种基于学习向量量化(LVQ)神经网络的人脸朝向识别算法。首先,介绍了传统的LVQ神经网络算法,并分析了算法存在的问题;然后,根据图像和视频中人脸的位置、朝向和旋转都不固定的特点,对LVQ神经网络算法进行改进,优化了该算法各层间的权值参数,取得了较高准确度的识别效果;最后,将该算法与误差反传(BP)神经网络法进行识别效果对比测试。仿真结果表明,该算法人脸朝向识别正确率可达99%以上,识别率与抗干扰性效果较好。To solve the problem about low accuracy of the common face orientation recognition algorithms,a face orientation recognition algorithm based on learning vector quantization(LVQ)neural network is proposed.Firstly,the classic LVQ neural network algorithm is introduced,and its problems are analyzed.Then,according to the features about unfixed position,orientation and rotation of faces in images and videos,the LVQ neural network algorithm is improved.The weight parameters of each layer of LVQ neural network algorithm are optimized,and higher accuracy of recognition effect is obtained.Finally,the recognition effect testing of the algorithm and the back propagation neural network algorithm are compared.Simulation result shows that the correct rate of face orientation recognition can achieve more than 99%,and the recognition rate and anti-jamming effect are better.

关 键 词：学习向量量化 神经网络 人脸朝向识别 

分 类 号：E917[军事] TP391.4[自动化与计算机技术—计算机应用技术]