基于学习向量量化神经网络的人脸朝向识别方法  

作　　者：冯洁琼[1] 卢焕章[1] 陈尚锋[1] 

机构地区：[1]国防科学技术大学电子科学与工程学院,湖南长沙410073

出　　处：《数字技术与应用》2016年第5期95-96,99,共3页Digital Technology & Application

摘　　要：针对传统人脸朝向识别算法中识别准确率较低的缺点,本文采用基于学习向量量化神经网络的识别方法,通过提取人脸图像中眼睛位置的特征向量并对朝向不同的人脸图像样本进行学习训练,优化了学习向量量化神经网络各层间的权值参数,取得了较高准确度的识别效果;仿真结果表明,采用学习向量量化神经网络的识别方法对人脸朝向进行识别可行有效,正确识别率可以达到95%以上,识别率与抗干扰性明显优于误差反传神经网络法。Aiming at the low accuracy disadvantage of traditional facial orientation recognition algorithm, the paper employs the recognition method based on the Learning Vector Quantization neural network. By means of extracting the feature vector of eyes positions in the face images and studying the different facial image samples, the paper optimizes the weighting parameters of the LVQ neural network, which achieves good recognition result. The simulation results indicates that the facial orientation recognition based on the learning vector quantization neural network is feasible and effective, and the correct recognition rate can reach more than 95%. Besides, the paper ultimately proofs the accuracy and the validity of the learning vector quantization neural network is better than the Back-Propagation neural network.

关 键 词：人脸朝向识别 学习向量量化 神经网络 特征向量提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]