人脸识别的面部特征配准及人脸比对问题研究
发布时间：2018-08-01 13:11
【摘要】：随着大数据时代的到来,个人与国家的信息安全正在逐渐成为一个研究热点。而生物识别技术因为其安全性、保密性及方便性等优点迅速成为了科研人员的“宠儿”。在众多的生物特征识别技术中,人脸识别技术以其无接触性、高效性、便捷性、唯一性、精准性等优点脱颖而出,发展成了研究热度最高的生物特征识别技术。通常的人脸识别系统中面部特征配准模块和特征提取与比对识别模块占有重要地位,本文针对这两个内容展开了深入研究,主要研究工作如下:首先,概述人脸识别的研究历史现状与基本技术方法;面部特征配准的研究历史现状与技术方法;人脸比对的研究现状、应用与发展方向。接着,研究人脸检测与人脸图像预处理环节。对当前存在的主要人脸检测方法进行了概述和分类。从特征的选择,强分类器的生成,级联检测器的构成详细讨论基于Haar_like特征与基于LBP特征的AdaBoost的人脸检测方法。通过对这两种方法的实时性与准确性的比较得出基于Haar_like特征的AdaBoost人脸检测方法具有较好的描述能力;基于LBP特征的AdaBoost人脸检测方法时效性比较好。在检测过后,通过尺度归—化和灰度变换统一人脸区域尺寸,消除颜色信息。然后,从两个方面对人脸面部特征配准方法进行了研究。一方面是基于几何特征,从人脸面部特征点出发,介绍了基于显式形状回归的面部特征配准方法。在不同的数据库进行配准实验,给出了比较全面的人脸配准效果图。另一方面是基于统计特征,研究基于不变形变换主成分分析的人脸配准方法,讨论了KL变换、特征空间的创建以及反向合成算法的迭代过程,用手动对齐的标准人脸库对该方法进行了实验验证,结果表明,该方法能比较好地配准人脸,并与识别有着相互促进的效果。接着,研究了相似度度量问题。通常的度量方法仅仅是考虑下了一对样本的差异性,为了增加判别性,同时考虑人脸样本的共性和个性,采用联合共性和个性的度量方法对人脸样本对进行相似度度量。并在不同数据库对该方法进行实验验证,结果表明该方法能去的满意的结果。最终将所以环节联系起来,构建一个人脸比对系统。
[Abstract]:With the arrival of big data era, the information security of individuals and countries is becoming a research hotspot. Because of its advantages of safety, confidentiality and convenience, biometric technology has quickly become the favorite of researchers. Among the many biometric recognition technologies, face recognition technology has become the most popular biometric recognition technology because of its advantages of non-contact, high efficiency, convenience, uniqueness, accuracy and so on. In the common face recognition system, the facial feature registration module and the feature extraction and comparison recognition module play an important role. In this paper, the two contents of the in-depth research, the main research work is as follows: first, This paper summarizes the history and basic technical methods of face recognition, the history and technical methods of facial feature registration, and the research status, application and development direction of face matching. Then, face detection and face image preprocessing are studied. The existing methods of face detection are summarized and classified. A face detection method based on Haar_like features and AdaBoost based on LBP features is discussed in detail from feature selection, generation of strong classifiers and construction of cascaded detectors. By comparing the real-time and accuracy of the two methods, it is concluded that the AdaBoost face detection method based on Haar_like features has better description ability, and the AdaBoost face detection method based on LBP features has better timeliness. After the detection, the face region size is unified by scale normalization and gray scale transformation, and the color information is eliminated. Then, face feature registration method is studied from two aspects. On the one hand, based on geometric features and facial feature points, a facial feature registration method based on explicit shape regression is introduced. In different database registration experiments, a more comprehensive face registration effect map is given. On the other hand, based on the statistical features, the face registration method based on the principal component analysis of the invariant transformation is studied, and the KL transform, the creation of the feature space and the iterative process of the inverse synthesis algorithm are discussed. The experimental results show that the proposed method can be used to match human faces well, and it is mutually beneficial to recognition. Then, the similarity measurement problem is studied. In order to increase the discriminability and to consider the commonness and individuality of face samples, the common measurement method is used to measure the similarity of face samples. The experimental results in different databases show that the method can get satisfactory results. Finally, the link will be linked to build a human face comparison system.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
本文编号：2157690
[Abstract]:With the arrival of big data era, the information security of individuals and countries is becoming a research hotspot. Because of its advantages of safety, confidentiality and convenience, biometric technology has quickly become the favorite of researchers. Among the many biometric recognition technologies, face recognition technology has become the most popular biometric recognition technology because of its advantages of non-contact, high efficiency, convenience, uniqueness, accuracy and so on. In the common face recognition system, the facial feature registration module and the feature extraction and comparison recognition module play an important role. In this paper, the two contents of the in-depth research, the main research work is as follows: first, This paper summarizes the history and basic technical methods of face recognition, the history and technical methods of facial feature registration, and the research status, application and development direction of face matching. Then, face detection and face image preprocessing are studied. The existing methods of face detection are summarized and classified. A face detection method based on Haar_like features and AdaBoost based on LBP features is discussed in detail from feature selection, generation of strong classifiers and construction of cascaded detectors. By comparing the real-time and accuracy of the two methods, it is concluded that the AdaBoost face detection method based on Haar_like features has better description ability, and the AdaBoost face detection method based on LBP features has better timeliness. After the detection, the face region size is unified by scale normalization and gray scale transformation, and the color information is eliminated. Then, face feature registration method is studied from two aspects. On the one hand, based on geometric features and facial feature points, a facial feature registration method based on explicit shape regression is introduced. In different database registration experiments, a more comprehensive face registration effect map is given. On the other hand, based on the statistical features, the face registration method based on the principal component analysis of the invariant transformation is studied, and the KL transform, the creation of the feature space and the iterative process of the inverse synthesis algorithm are discussed. The experimental results show that the proposed method can be used to match human faces well, and it is mutually beneficial to recognition. Then, the similarity measurement problem is studied. In order to increase the discriminability and to consider the commonness and individuality of face samples, the common measurement method is used to measure the similarity of face samples. The experimental results in different databases show that the method can get satisfactory results. Finally, the link will be linked to build a human face comparison system.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
【参考文献】
相关期刊论文 前6条
1 阮晓虎;李卫军;覃鸿;董肖莉;张丽萍;;一种基于特征匹配的人脸配准判断方法[J];智能系统学报;2015年01期
2 江伟坚;郭躬德;赖智铭;;基于新Haar-like特征的Adaboost人脸检测算法[J];山东大学学报(工学版);2014年02期
3 王振奇;;人脸识别技术在安防行业的飞跃与突破——访汉王科技股份有限公司大客户事业部副总经理石践[J];中国安防;2010年08期
4 武京伟;黄春庆;;一种基于改进弹性束图匹配的人脸识别[J];工业控制计算机;2009年09期
5 李丽娟;石红伟;王爱云;;一种基于贝叶斯切线模型的人脸对齐算法[J];计算机工程;2009年14期
6 山世光,高文,唱轶钲,曹波,陈熙霖;人脸识别中的“误配准灾难”问题研究[J];计算机学报;2005年05期
相关硕士学位论文 前6条
1 郭志芳;基于支持向量回归的面部特征点定位算法[D];南昌航空大学;2015年
2 靳一凡;基于级联卷积神经网络的人脸关键点检测算法[D];浙江大学;2015年
3 江伟;人脸面部特征点检测及其在视频监控中的应用[D];上海大学;2015年
4 蔡超;基于面部基准点对齐的人脸识别方法研究[D];华中科技大学;2013年
5 孟繁特;人脸识别关键技术研究[D];哈尔滨工程大学;2012年
6 吴证;人脸特征点定位研究及应用[D];上海交通大学;2007年
，本文编号：2157690
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2157690.html
