【作者】 吴证；

【导师】 周越；

【作者基本信息】 上海交通大学 ， 模式识别与智能系统， 2007， 硕士

【摘要】 随着科技的进步和人们安全意识的提高,对快速有效的身份鉴别的实际需求日益迫切。人脸相比其他人体生物特征具有直接、友好、方便的特点,因此自动人脸识别成为身份鉴别的研究热点,有着广泛的应用前景。一个自动人脸识别系统一般包括:人脸检测、关键点特征定位、识别等步骤。影响自动人脸识别系统真正实用化的主要因素有:姿态问题、光照问题、表情问题等。特征点精确定位是解决这些问题的关键。本论文对于人脸识别中的人脸特征点定位、人脸特征点跟踪以及相关应用进行了深入研究,并在此基础上对相关算法进行了创新性的改进。论文的主要工作如下:1)系统的综述了人脸特征点定位的发展历史和研究现状。详细总结了基于灰度信息、基于先验规则、基于几何形状、基于统计、基于小波和小波包等人脸特征点定位方法,并分析和比较各种方法的优缺点。2)对人脸特殊器官的定位进行了研究。首先介绍了霍夫变换法、变形模板法、边缘特征分析法、对称变换法和基于彩色图像色度经验公式法等定位眼睛的方法。然后研究了通过聚类定位嘴巴的方法。3)介绍了基于模型的两种经典的特征点定位算法:主动形状模型(ASM)和主动外观模型(AAM)。并在本文的实验中构造了主动形状模型和主动表观模型,通过理论地分析和实验对ASM和AAM算法进行比较,提出改进方案。4)具体地对主动形状模型(ASM)算法进行了创新性地改进,提出了一种在彩色图像中结合肤色概率信息的改进ASM算法,并进一步提出了基于人脸特征点Gabor小波特征分类的特征点搜索方法,对改进ASM的结果进行精确校正,达到鲁棒精确地定位特征点的目的。5)对光流分析算法进行了研究,并将该算法应用于人脸图像序列的特征点的跟踪,并通过实验证明了其有效性。6)本文对基于特征点跟踪的唇形识别进行了探索性研究,提出了一种特征点位置和运动特征提取的方法,并介绍了基于子空间学习的几种经典的特征降维方法:主元分析法(PCA)、线性判别分析(LDA)、基于流形学习的降维(LPP)、辨别共同向量(DCV),最后通过实验对这几种降维方法的识别效果进行比较,从而证明本文提出的基于特征点跟踪的唇形识别算法的可行性。更多还原

【Abstract】 With the development of society and technology, the demand for effective automatic verification is increasingly urgent. Comparing with other human biological characteristics, face characteristics have direct, friendly and convenient characteristics, therefore automatic human face recognition becomes the research focus of identity verification and has extensive application foreground. General speaking, an automatic face recognition system contains: face detection, feature points extraction and recognition. There are some factors which affect instantiation of face recognition. They lie on pose, light and expression of face. Precise facial feature points extraction is the key to solve these problems.In this dissertation, algorithms of facial feature points extraction, tracking and relevant application were profoundly researched. Based on the classical algorithms, novel and creative improvement was proposed. The main discussion of the dissertation is listed as follows:1) The history and status of research on facial feature points extraction are systematically summarized. Detail kinds of feature points extraction approaches, including grey-level-based algorithm, knowledge-based algorithm, Geometry-base algorithm, statistic-based algorithm, wavelet-base algorithm, and then analysis the different algorithm.2) The algorithms of locating rough position of special facial organs are researched. Introduce eyes locating methods including Hough Transform algorithm, Deformable Template algorithm, edge feature analysis algorithm and knowledge-based algorithm. And then introduce Clustering-based mouth locating method.3) Introduce two classical algorithms of facial feature points extraction: Active Shape Model (ASM) and Active Appearance Model (AAM). In this dissertation, we construct ASM and AAM for experiments. With comparing the two algorithms not only in theory but also in experiments, we conclude the weakness and propose the improving solution.4) A method of facial feature points extraction based on improved Active Shape Model and Gabor wavelet features is presented. And the experiments prove that facial feature points can be located robustly and precisely using the method proposed.5) Research the optical flow algorithm and apply it to feature points tracking in serial images of face.6) Explore the method of lip-movement recognition based on track of mouth feature points. An algorithm that extracts the features of lip points’ track is proposed. Then introduce subspace learning based dimension reduction algorithm: PCA, LDA, LPP, DCV. And the experiments prove the feasibility of this algorithm proposed.更多还原