【作者】 许高凤；

【导师】 丁士圻；

【作者基本信息】 哈尔滨工程大学 ， 通信与信息系统， 2009， 博士

【摘要】 随着社会的进步以及各方面对于快速有效的自动身份认证的迫切需求,生物特征识别技术在近几十年中得到了飞速的发展。生物特征作为人的一种内在属性,具有很强的自身稳定性及个体差异性而成为了自动身份认证的最理想依据。人脸识别作为生物特征识别的一种,由于具有直接、友好、方便等特点,得到了广泛的研究与应用。自从上世纪六十年代Chan & Bledsoe提出利用人脸特征进行身份识别与鉴定以来,人脸识别技术得到了长足的发展,并出现了一些典型算法与系统。在理想条件下,这些算法均能取得较好的识别效果,但实际测试结果表明,一旦条件发生变化、不再理想时,其识别率将大打折扣甚至系统不可用。影响人脸识别率下降的因素诸多,主要有光照、表情、年龄、姿态等。因此要研究出实用的、大众可接受的人脸识别系统还有很多工作要做。本文是在现有人脸识别技术的基础上,重点探讨了影响人脸识别性能的几个方面。本论文的主要工作如下：(1)研究了人脸特征点的精确定位方法,介绍了常用的人眼定位方法,指出了这些方法的不足,并提出了一种联合的人眼睛精确定位算法,并测试、分析了该算法定位效果。(2)总结了人脸识别中现有的光照问题解决方法,利用小波技术分析了光照、表情等变化对人脸图像低频信息的影响。提出了一种基于小波重建的人脸去光照方法,并分析和比较了该算法处理受光照、表情影响人脸图像的识别效果。(3)分析了现阶段比较典型的人脸图像的特征提取方法,指出了各个算法的优点和不足之处,提出了一种新提取人脸图像中有利于识别的全局信息和局部结构信息的独立源特征的方法。并在公开数据库上进行了测试比较与分析。(4)设计并实现了一个模块化的人脸识别算法仿真与开发平台,并详细地介绍了各个模块应用的算法与工作流程。结合人脸识别技术在大型场馆门禁系统中的应用,给出了一种基于集群计算机工作模式的门禁系统,并引入了Intel的MMX\SSE技术的系统并行加速算法。(5)介绍了人脸识别门禁系统在2008年好运北京测试赛以及29届北京奥运会场馆门禁系统中的应用,给出了系统测试性能,并进行了分析。本论文的研究成果不但在鲁棒人脸识别技术上有一定的参考意义,而且对系统的高速实时运行有一定的借鉴意义,所提出的算法部分已成功地应用在2008年好运北京测试赛、29届北京奥运会以及残奥会场馆门禁系统中。更多还原

【Abstract】 Biometric technology has made rapid development in recent decades since the importance of public safety. The biological characteristics are inherent attributes. They have strong self-stability of individual differences, and are automatically the best evidences for authentication. Face Recognition as one of the biometric technologies, is direct, friendly and convenient. It has been widely studied and applied.Since the 1960s Chan & Bledsoe provided using human face for identification, face recognition technology has developed by leaps and bounds, and there have been some typical algorithms and systems. In ideal conditions, these algorithms can make better recognition results, but the actual test results show that, once the conditions change, which are not ideal, its recognition rate will be greatly reduced. The factors which affect the recognition rate are mainly light, expression, age, posture, etc. Therefore, there has a lot of work to do for practical and acceptable public face recognition technology. Based of the existing face recognition technologies, this paper focuses on several key aspects of performance. The main contribution of this paper is summarized as follows:(1) Facial key point location, in particular the exact position of the eyes is the premise of face image cropping and regulation. This paper presents the existing location technologies and points put their shortcomings. Then it presents a joint eyes precise positioning method, and gives the test results and the analysis.(2) Since the illumination has great impact on face recognition, and it usually changes in the low frequency domain, the paper analyses its impaction on image low-frequency coefficients of multi-wavelet decomposition. After that a low-frequency coefficients reconstruction method, which based on the PCA, is proposed. And a full test which based on the classic Eigenface is provided.(3) This paper presents an effectively feature extraction method. It first processes the face image by the LFA (Local Feature Analysis), and gets the LFA feature. After that we decompose the feature by ICA (Individual Component Analysis) method, and use the genetic algorithm (GA) on the decomposed signal for selecting the proposal signal for recognition.(4). In view of the continuous development and updating of face recognition technology, and the new emergence of the large number of new algorithms, we construct a system with the combination of several algorithm blocks as a platform for new algorithm development and simulation.(5). In accordance with the existing face recognition technology and the application in venues access control system, this paper introduce a access control system based on the computer cluster. And according to the requirements of practical application, the system adopts the client\server (C\S) mode. At the same time in order to achieve real-time process rapidly on large database, it introduces a parallel speedup algorithm based on the Intel’s MMX\SSE technology. In the last, of this paper, it presents the venues access control system of Good Luck Beijing Sports and the 29th Beijing Olympic Games, and gives the performance of the recognition system.The results of paper not only can be some references in the technical sense of robust face recognition, but also can be a reference for high-speed real-time operating system, a number of algorithms have been successfully applied to the existing our system.更多还原