摘要
针对三维人脸特征表示问题,提出了一种三维人脸轮廓曲线特征与二维Gabor小波特征相融合的人脸特征表示新方法。基于这种新的融合特征,利用模糊自适应共振神经网络(Fuzzy ARTMAP)进行有监督的学习训练,并构建三维人脸识别分类器实现人脸识别。利用模糊自适应共振神经网络分类器的增量学习能力,可以很好地解决随着训练模型增加导致识别系统识别率降低的问题。所提出的方法在FRGC v2.0三维人脸数据库上,对人脸表情变化进行了实验。结果表明具有一定的鲁棒性,识别率高,且随着新增人脸数量的增长可以提高准确率。
This paper presents a new representation of 3D face recognition features, includeing the 3D face curve features and 2D Gabor wavelet features. Based on the new multi-features, Fuzzy ARTMAP is used to train the network with supervised learning, and create a 3D face recognition classifier to distinguish human faces. The algorithm can be used to keep a high recognition rate with the incremental number of models. The experiment test on FRGC v2.0 database shows that the performance of multi-features fusion is perfect on robustness in changing expressions, and Fuzzy ARTMAP is used to keep a high recognition rate in the increasing number of human faces.
出处
《机械制造与自动化》
2015年第4期126-131,共6页
Machine Building & Automation
基金
福建省自然科学基金资助(2013J01226)
