【作者】 皇甫征声；

【导师】 汪同庆；

【作者基本信息】 重庆大学 ， 模式识别与智能系统， 2003， 硕士

【摘要】 人体生物特征识别技术(Biometrics)，又称生物测定学，是使用人体本身固有的物理特征（如指纹、虹膜、面部、掌纹等）及行为特征（如书写、声音、击键等），通过模式识别的方法来自动鉴别个人身份的技术，与传统方法相比，具有更好的安全性、可靠性和有效性。人脸作为人类视觉鉴别中最为普遍的模式，是一种最容易被接受的身份鉴定方法，因此人脸识别也成为最有潜力的生物身份验证手段之一。本文收集和分析了大量近年来国内外关于人脸检测与人脸识别的学术论文及研究报告，针对人脸自动识别系统的建立进行了深入的研究，运用信息融合的思想，提出了一种用于视频图像正面人脸检测与人脸识别的方法。实验证明本文提出的方法是合理的，有一定的理论价值与实用价值。本文的研究工作主要包括以下几个方面：1、通过对国内外在人脸检测方面研究成果的分析，针对动态视频图像的特点，提出了综合采用基于运动检测的背景分离技术以及基于肤色和对称性特征融合检测技术的方法、马赛克检测方法和神经网络方法的决策层融合方法。该方法实现了基于知识方法和基于统计方法的综合，实现了特征层和决策层的多层融合，其决策融合采用混联的结构，兼顾了速度和检测率。实验证明本文的检测方法是合理的、高效的。2、在对各种人脸识别方法进行分析的基础上，重点研究了基于隐马尔可夫模型（HMM）的人脸识别方法，进而提出了一种将人脸图像进行小波分解，然后利用分解后的代表原始图像不同细节特征的四幅小波子图像分别建立HMM进行学习，然后将四个模型进行决策融合的方法，这种方法比单个HMM更好地反映了人脸的统计特征，得到了更好的识别效果。采用基于最佳鉴别矢量集的方法解决HMM分类器难以实现拒识的问题。多个人脸数据库的测试证明了该方法的有效性。3、对信息融合的理论和方法进行了研究，分析了分类器融合的速度问题，采用置信分析的方法寻找速度和识别率的最佳折中，分析了决策融合的经典投票方法进而给出改进算法，使多分类器决策融合的识别率得到进一步提高。实验证明信息融合方法能提高检测识别系统的识别率和鲁棒性。4、对基于生物识别技术的应用系统进行了研究，结合以本文人脸检测识别算法为核心的企业人事考勤管理系统，分析了系统设计中的软件工程方法，并简要介<WP=5>绍了该系统的方案设计和数据库设计。更多还原

【Abstract】 Biometrics, a kind of technology in personal identification system, is the totally brand-new, safe, reliable and effective technique, that is different from traditional method, because it use physical and behavior characteristic of human body such as fingerprint, iris, face, handprint, writing or voice. Human face is the most popular pattern and the most easily accepting method in the daily life, so person’s face identification is becoming the most promising one of the verification methods. Learned the domestic and international discourse and research papers concerning face detection and recognition in recent years, some theories problem to computer identify technique to person’s face is analyzed. Aim at establishing the personal identification system for mobile video; using the idea of information fusion, the dissertation brings forward a face detection and recognition method. Experiments prove the method in this dissertation is reasonable, and have great value in theories and practice. The textual main research work primarily includes:1. Via domestic and international research productions at face detection, the dissertation bring forward a detection method aim at mobile video, in which back-ground separating based movement detection, color and Symmetry based features fusion, mosaic and neural network are used synthetically. In this method, knowledge based method and statistic based method is synthetically. This method realizes information fusion on feature level and decision level in which mixed structure that can split the difference of speed and recognition rate is used. Experiments prove the method in this dissertation is reasonable, effective.2. On the base of research on multifarious methods, the dissertation studied face recognition method based on hidden Markov model (HMM). On foundation of some improvement in HMM method, the dissertation first decompose face image to four son images by wavelet transform,second make use of the decomposition images to establish HMM modules respective, at the last acquire the result use decision fusion. Because of better expression of statistical features, this method acquire better recognition rate. The dissertation use optimal discriminate vectors based method to solve the reject rate problem HMM based method. The test on some face database proves that this method can increase recognition rate.<WP=7>3. On the base of research on information fusion, the dissertation analyze the speed of multi-classifier and bring forward the method that can split the difference of speed and recognition rate, believe analysis. The dissertation analyzes classical vote method in decision fusion and brings forward improved method. Experiments prove that the method can heighten the recognition rate and robust.4. With the research on Biometrics based application system and application design experiences of mine, the dissertation analyze corporation human resource and check system design process, software engineering, blue print design and database design is mainly discussed.更多还原