基于改进Gamma和改进BP算法的人脸识别研究  

Face recognition based on improved Gamma and improved BP algorithm

作　　者：李国芳[1] 王力[1] 

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025

出　　处：《微型机与应用》2015年第4期49-51,共3页Microcomputer & Its Applications

摘　　要：针对传统BP算法收敛缓慢、训练过程振荡等缺点,提出了一种基于改进Gamma和改进BP神经网络的人脸识别算法。采用改进Gamma矫正方法改善原始图像的光照不均匀,并采用小波变换和NMF算法提取图像主要特征,最后结合一种新的权值调整方法改进BP算法进行图像分类识别。仿真实验表明,与传统算法相比,使用该算法对Yale人脸库、CMU PIE人脸库和自建人脸库的图像进行识别具有更快的收敛速度和更高的识别率。In light of the deficiencies including slow convergence and training process-oscillation of traditional BP neural network,this paper proposes a face recognition algorithm based on NMF and improved BP neural network. It improves the uneven illumination of the original image with the improved Gamma correction method, combines wavelet transform with NMF algorithm to extract image features, and uses a new weight adjustment method to improve the BP algorithm for image classification and recognition. Simulation results show that faster convergence speed and higher recognition rate are achieved when using the improved algorithm to identify the images in Yale face database, CMU PIE and self-buih face database than the traditional algorithm.

关 键 词：人脸识别 小波变换 GAMMA 非负矩阵分解 BP神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]