【作者】 韩文静；

【导师】 李晶； 曹茂永；

【作者基本信息】 山东科技大学 ， 模式识别与智能系统， 2009， 硕士

【摘要】 随着信息技术的发展和日益增长的对安全技术的需要,基于生物特征的身份识别技术在近年来有了迅速发展。作为生物特征识别技术之一的人脸识别技术正在兴起,并显示了很大的优越性。在现有的各类生物识别技术中,人脸识别技术具有使用方便、用户接受度较高、具有直观性、不易被他人仿冒、人脸图像易于获得等优点。本文以CVL人脸图像数据库为基础,研究了人脸识别的典型算法,并对某些方法进行了改进,最后实现了一个基本的人脸识别过程,本文的主要工作如下:第一,人脸图像的预处理。主要进行了灰度化、粗切、图像平滑、图像增强及灰度归一化等预处理操作,消除了图像噪声,使图像更清晰。第二,人脸的检测与定位。首先利用垂直灰度投影曲线得到人脸的轮廓图像,再利用基于混合投影函数(HPF)和Hough变换的方法进行眼睛的精确定位,并以眼睛为基准进行人脸图像的校准,包括旋转、剪裁和缩放等,最后图像变为100×100的标准图像。第三,特征的选择和提取。主要研究了主成分分析(PCA)、2DPCA和独立分量分析(ICA)的算法,然后对2DPCA的算法进行了实验验证与分析,并与PCA的方法进行了比较。最后,分类决策。主要研究了最近邻法则、BP神经网络的方法及MCFD的算法,并选择MCFD的算法进行本文的分类器设计。实验结果表明,2DPCA和MCFD相结合的算法能够得到较好的识别结果和较快的识别速度。更多还原

【Abstract】 With the development of information technology, biometric recognition technologies have been developed rapidly in recent years. Face recognition is relatively new among the biometric recognition technologies because of its huge potential superiority. This method shows great advantages such as convenient operation, higher acceptance, and stronger visual impression.A face recognition system based on CVL database was realized, in which the algorithms of two-dimensional principal component analysis (2DPCA) and multi-classifiers fusion decision (MCFD) is developed. The major work includes:Firstly, image preprocessing algorithms were studied. By image smoothing, enhancement and gray-scale normalization, the noise was cancelled and the influence of illumination was wiped off, thus cleaner images are obtained.Secondly, image detection and orientation was studied. After the outlines of face images were obtained by the vertical grayscale projection function, the Hybrid Projection Function (HPF) and Hough transform were used to decide the orientation of eyes, then, the images were standardized from the size of 640 x 480 to the size of 100×100.Then, feature selection and extraction algorithms were studied, including principal component analysis (PCA), 2DPCA and indpendent component analysis (ICA), based on which 2DPCA was used compared to PCA, better performance was achieved.Finally, the design of classifier was studied. The algorithms such as the nearest neighbor method, BP neural network, and MCFD are studied, and then MCFD was adopted.Through experiments, the algorithm based on 2DPCA and MCFD was adopted for its higher recognition rate and faster operating speed.更多还原