【作者】 洪玥；

【导师】 郎利影；

【作者基本信息】 河北工程大学 ， 计算机应用技术， 2009， 硕士

【摘要】 人脸识别是当前计算机模式识别领域中图像分析和理解领域中的非常重要的课题,赋予了计算机识别人物身份的能力,涉及了模式识别、计算机图形学、数字图像处理等热门研究领域。与利用指纹、虹膜等其他人体生物特征进行身份识别的方法相比,人脸识别更加友好、方便和隐蔽,在公安追逃系统、安全验证系统、档案管理系统、信用卡验证、视频会议、人机交互系统等方面都具有重要的应用价值。虽然人类能毫不费力地识别出人脸及其表情,但人脸的自动机器识别却是一个难度极大的课题,牵涉到计算机视觉、模式识别、生理学和心理学等方面的诸多知识。本文对人脸识别问题的现有理论和算法进行了探讨、研究,在此基础上,针对某些环节提出了改进算法,主要研究内容如下:1.研究并改进了基于Fisher鉴别向量的人脸识别技术。首先,研究了Fisher最佳鉴别向量、最佳鉴别平面和多类问题的Fisher线性判别;在此基础上,详细讨论了一种基于PCA方法和Fisher线性判别的Fisherface方法,并且给出了采用Fisherface方法进行识别的实验结果。2.提出了基于核Fisher判决分析的人脸识别方法。该方法的核函数采用了修正模型和分数次幂多项式核函数,在一定程度上解决因光照、姿态等因素的适量变化产生的识别率下降的问题。通过在具有多表情、多姿态的人脸库上进行仿真实验,实验结果表明该技术相对于现有其它基于核函数的分析方法,具有更高的识别率和更强的鲁棒性。3.采用模式特征之间的欧式距离作为相似性度量构成最小距离分类器,将未知的人脸图像在预先训练好的人脸库中进行比对,从而完成对未知人脸的识别。4.设计了一个基于静态的人脸识别系统,该系统能满足一定的识别速度和准确率要求。更多还原

【Abstract】 The face recognition is a very important subject in the area of image analysis and comprehension of computer pattern recognition. It endows the computer with the ability of recognizing person identity, which relates to some hot research fields such as pattern recognition, computer graphics, digital image processing and so on. Compare with other biometric technologies, such as finger print recognition and iris recognition, face recognition has characteristics of non-contact and friendly interface. Face recognition has great significance of application in these fields such as public security tracing system, security validating system, files-management system, credit card testing system, video conference system and human machine communication system. Although human can recognise face and its expression with any efforts, face recognition is still a difficult task for computer, which is closed related to computer vision, pattern recognition, physiology, psychology and etc.To the problems of face recognition, respectively, this dissertation reviews the existing theory and algorithm. Furthermore, it proves the old approaches. The main content of this dissertation is summarized below:1. The Fisher discriminate vector based face recognition technique is worked over. First, we study the Fisher best discriminate vector, and best discriminate plane and the Fisher linear discriminate rules for multrclasses. Then based on this, we discuss a PCA based method and a Fisher linear discriminate method in details. We gave the Fisher base face recognition method and the corresponding experiment results.2. In order to solve recognition rate reduction resulting from illumination, pose etc moderate variations to some extent, the Kernel Fisher Discriminant Analysis method is studied and a corrected model and a fractional power polynomial kernel function are employed in the method. Through experiments on face databases with multi-poses and multi-illumination conditions, the results indicate that in contrast to other methods based on Kernel, the method is of higher recognition rate and more robustness to the above variations.3. In the matching stage, the traditional minimum distance classificacion is improved to recognize the unknown faces by using Euclidean distance as the similarity measurement. 4. A face recognition system based on the static is designed, which can satisfy, to some extent, the requirements for recognizing spend and accuracy.更多还原