基于BP神经网络的人脸识别系统研究
2018-09-05陈翔白创黄跃俊
陈翔 白创 黄跃俊



文章编号: 2095-2163(2018)03-0057-04中图分类号: 文献标志码: A
摘要: 关键词: (School of Physical & Electronic Science, Changsha University of Science & Technology, Changsha 410114, China)
Abstract: A face recognition system based on BP neural network is designed and its performance is analyzed in this paper. This system first uses discrete wavelet transform to obtain low-frequency components that contain most of the original information of the face image, and reduces the dimension of the image data. Then the PCA algorithm extracts the principal component features of the face image to further reduce the amount of image data processing; The trained face of the BP neural network is used to classify the face to be measured. Discrete wavelet transform, PCA feature extraction and BP neural network classification design are introduced in detail. Through system simulation experiments and analysis, it is found that the extraction of facial features is the key to this system; due to statistical characteristics of face gray information and the BP neural network classification by supervised training, the system is only to be used in the face recognition application scenario with a high recognition accuracy, in which the category is fixed, and the light is uniform . So it is applied in complex environments difficultly.
Key words:
基金项目: 作者简介:
通讯作者: 收稿日期: 引言
人脸识别技术是目前人工智能与机器视觉研究领域的重点技术之一,有着广阔的应用前景,如摄像头监控、刷脸支付及身份识别等。人脸识别的研究可前溯至于上世纪60年代末。早期的人脸识别方法主要分为2种:基于人脸几何特征法和模板匹配法。前者是将人脸图像进行归一化,并求取各特征点之间的距离、比率,再对特征点所构成的二维拓扑结构进行识别;后者则是通过计算人脸模板与待测人脸图像灰度值之间的相关性来划定分类\[1\]。90年代后计算机技术得到了快速的发展,对更加精确的人脸识别方法的需求也在增多,基于神经网络的人脸识别方法即已成为该领域的关键研发内容\[2\]。
基于神经网络的人脸识别方法能有效地利用神经网络的并行分布式处理及高速寻找最优解的能力,大大减少了以往的图像数据的运算时间。同时相比于其它人脸识别方法,神经网络的优势特点也更加鲜明,因其可通过训练学习的过程来获得人脸识别潜在规律的隐性表达,适应能力强,识别率也高\[3\]。
本文采用基于误差反向传播算法的BP神经网络的人脸识别方法设计人脸识别系统。该系统中,为提高BP神经网络的识别速率和准确率,采用PCA分析法对人脸图像经离散小波变换后获得的低频分量,进行降维并提取主成分特征。……
