【作者】 赵晓晴； 王娟娟； 毋立芳；

【Author】 Zhao Xiaoqing Wang Juanjuan Wu Lifang (School of Electronic Information & Control Engineering,Beijing University of Technology,Beijing 100022)

【机构】 北京工业大学电子信息与控制工程学院；

【摘要】 本文研究并实现了小波变换,光照商图像与BP神经网络相结合对不同光照条件下的人脸图像进行识别的算法。运用小波变换方法对图像进行降维,运用光照商图像方法对图像进行预去光,运用BP神经网络进行人脸识别;同时讨论了BP网络输入矢量的标准化处理问题、网络隐含层神经元数选取问题,以及对Sigmoid函数和网络学习速率的改进问题。利用Yale Face Database B人脸数据库进行识别实验,结果表明,通过预去光,网络的识别率有了显著提高;本文设计的改进BP网络在与非改进网络得到相同识别率情况下,其收敛速度明显加快。更多还原

【Abstract】 In this paper,face recognition under different illumination conditions based on Wavelet transform,Illumination ratio image and BP neural network has been studied and implemented.Wavelet transform is applied to reduce the dimension of the images,illumination ratio image is applied to conduct illumination alignment,BP neural network is applied to conduct face recognition.Also discussed are the problem of normalization of the input vector,the problem of determining the number of hidden layer’s neural nodes,and Sigmoid function and training speed of BP neural network are improved.Yale Face Database B is used to conduct experiments,the recognition results show that the recognition rate has been improved markedly by illumination alignment; the improved BP neural network has better convergence speed than non- improved BP neural network.更多还原