一种基于改进BP神经网络的PCA人脸识别算法  

PCA FACE RECOGNITION ALGORITHM BASED ON IMPROVED BP NEURAL NETWORK

作　　者：李康顺[1,2] 李凯[1] 张文生[3] 

机构地区：[1]江西理工大学理学院,江西赣州341000 [2]华南农业大学信息学院,广东广州510642 [3]中国科学院自动化研究所,北京100190

出　　处：《计算机应用与软件》2014年第1期158-161,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(70971043);江西省教育厅科学技术研究项目(GJJ112348)

摘　　要：人脸识别作为模式识别领域的热点研究问题受到了广泛的关注。传统BP算法虽然具有自学习、自适应以及强大的非线性映射能力并且在人脸图像识别准确率上占有很大的优势,但算法具有收敛缓慢、训练过程振荡、易陷入局部极小点等缺点。针对传统BP算法的不足提出一种基于改进BP神经网络的PCA人脸识别算法,该算法采用PCA算法提取图像的主要特征,并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,通过使用该算法对ORL人脸数据库的图像进行识别,其结果比传统算法具有更快的收敛速度和更高的识别率。Face recognition, as a focus of the research in pattern recognition field, has gained increasing attention. Traditional BP algorithm has a strong ability in self-learning, self-adaptivity and nonlinear mapping. Moreover, it has a significant predominance in human face recognition accuracy. However, the algorithm also has shortages including slow convergence, training process oscillation and easy to fall into local minima. In light of these deficiencies of traditional BP neural network, we propose a PCA face recognition algorithm which is based on improved BP neural network. The algorithm uses PCA algorithm to extract principal features of face image and uses a new weight adjustment method to improve the BP algorithm for image classification and recognition. Simulation experimental results show that faster convergence speed and higher recognition rate are achieved when using the improved algorithm to identify the images in ORL face database than the traditional algorithm.

关 键 词：人脸识别 主成分分析rBP神经网络 附加动量 弹性梯度下降法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]