人脸特征提取理论及其应用的的研究.docVIP

摘  要  人脸特征提取理论及其应用的研究  摘  要  近些年来，随着科学技术的发展和社会的进步，人们对自动、快速、有效的身份 鉴别的要求日益迫切。人脸作为主要的面部特征，具有直接、友好、方便且信息量大 等特点，成为最直接的身份鉴别手段。特征提取是人脸识别的重要步骤，特征提取的 好坏将直接影响到分类的效果。本文主要研究了人脸图像特征提取的方法，主要内容 如下： 1.介绍了人脸特征提取的常用方法：主分量分析、独立分量分析、线性鉴别法、 Fisherface 方法、统计不相关方法及核空间的特征提取方法，探讨了不同方法的优缺 点并给出了详细的实验结果与性能分析。 2.把模糊方法引入到人脸特征提取中，计算模糊情况下类内散布矩阵和类间散布 矩阵，从而获得更佳的投影方向，实验表明了该方法的有效性。 3.受万有引力定律的启发，我们建立了基于万有引力定律的人脸识别模型。首先 用 PCA 对图像进行特征提取,然后建立图像与图像之间的相似度模型以及类别对图像 潜在作用力的表达式,该作用力大小决定图像所属的类别。在 ORL 上的实验表明,该模 型有一定的可行性。 4.介绍通过光场来识别姿态变化的人脸的方法，首先把图像转化为光场中的向 量，用 PCA 或 LDA 等方法提取光场的特征向量，然后计算训练图像和测试图像在光场 特征空间的投影系数，用最小距离分类器进行分类，实验验证了该方法的优越性，特 别是在姿态变化或人脸非正面情况下。 关键词：特征提取，主分量分析，独立分量分析，万有引力定律，光场 -I- ABSTRACT Theory of Feature Extraction of Faces and Its Applications Abstract Recently, with the development of science and technology,face recognition become more and more important. Faces,which have the advantages of immediacy, amity, convenience,become the direct tool for face recognition.Feature extraction is the important part in the process of identification, and the result of feature extraction will affect classification.This dissertation mainly discusses the theory and application of feature extraction . The main contributions are summarized as follows: 1.The fundamental framework and typical methods of feature extraction have been intraduced including principal component analysis, independent component analysis, linear discriminant analysis,fisherface ，statistically uncorrelated discriminant vectors and kernel methods .The advantages and disadvantages of different methods, and the results of experiments are presented respectively. 2.The fuzzy technology is incorporated into feature extraction of faces. The between-class and within-class scatter matrices are reformulated. It is confirmed by experiments of face recognition that the fuzzy method is effective. 3.A model of face recognition based on the law of universal gravitation is established. PCA is performed to extract the features of face image. Then the si