基于人脸整体特征的证—人判别方法研究

【作者】 韩家新；

【导师】 何华灿；

【作者基本信息】 西北工业大学 ， 计算机软件与理论， 2002， 硕士

【摘要】 人像识别技术是一个极其复杂和困难的研究课题，它在网络安全、视频会议、人机智能交互等方面有着巨大的应用前景，因而人像识别技术成为当前模式识别和人工智能领域的一个研究热点。近二十年来涌现了各种不同的人脸识别方法和系统，但是并没有一个鲁棒性好，识别率高的识别系统在实际中得到应用。本文分析了这些方法的差异性，建立了一个基于子空间法的人脸识别方法研究环境。重点研究了基于PCA的人脸识别算法和基于Fisher线性判别的人脸识别算法，并将这两种子空间法合并为一个识别算法，对构成特征空间的特征向量个数的选择和不同的相似性度量方法进行了讨论，并提出了扩展维数法选择向量个数的方法。在实验的基础上比较了不同的识别方法和不同的度量方法对识别率的的影响，得出了一些有价值的结论。另外，本文还提出了基于仿射模板的人脸定位算法，实验结果证实了该算法的有效性。更多还原

【Abstract】 Face recognition is a complex and difficult problem that is important for surveillance and security, telecommunications, digital libraries, and human-computer intelligent interactions,so it is one of the most active research field on Al and pattern recognition.Over the last twenty years different method have been proposed,but there is not a system in application which is robust and high in recognition rate.The difference of these methods is analysized in the paper and a test environment for face recognition is put up.The paper focuses on two basic methods , one is based on PCA algorithms and the other is based on Fisher discriminants These two subspace face recognition algorithms are combined into one in the paper, moreover,the number of eigenvectors used to create the eigenspace and the behavior of similarity measures are discussed and one method called stretching dimension is given to select eigenvectors.experiments are presented comparing recognition rates for different algorithms and different similarity measures. In addition, A method of face location based on affine template matching is presented,and experimental results show the method has goodperformance.更多还原