摘要
研究了基于连续小波变换的神经网络进行人脸识别的方法.介绍了小波分析的理论基础,详细讨论了根据小波变换系数的范数选取小波母函数的方法,根据小波脊线确定网络神经元个数的方法以及神经网络的初始化和参数训练方法.通过对人脸图像灰度的连续小波分析,神经网络的自组织自学习能力,调整连接权值和小波神经元的尺度、位移参数,完成人脸识别的任务.实验结果验证了该神经网络的识别性能明显优于用特征脸方法对相同人脸库进行的识别.
A face recognition method using neural network is proposed based on continuous wavelet transform. The theory of wavelet analysis is introduced. The method of selecting mother wavelet function is detailed according wavelet transform norm, and the number of wavelet knot and the initialization are decided by wavelet ridge. The training method of parameters is also introduced. Through continuous wavelet analysis and self-organizing and self-training of neural networks, the connection weights and the parameters of scales and positions are adapted to complete the task of face recognition. The test results shows that this method obtains much better performance in face recognition compared with the classical eigenface method.
出处
《光子学报》
EI
CAS
CSCD
北大核心
2005年第9期1425-1430,共6页
Acta Photonica Sinica
关键词
连续小渡变换
人脸识别
模极值
小渡脊线
Continuous wavelet transform
Face recognition
Module maxima
Wavelet ridge