【作者】 蒲东兵；

【导师】 周春光；

【作者基本信息】 吉林大学 ， 计算机应用技术， 2009， 博士

【摘要】 二十一世纪的一个显著特征就是身份信息的数字化和隐性化,在这个网络化、信息化的时代,如何准确鉴定一个人的身份,保护信息安全已经成为必须解决的一个关键社会问题。传统的身份鉴别技术已经不能完全满足现代社会经济活动与发展的需要,采用人体自身所固有的生物特征来鉴别身份被认为是本世纪最安全的身份鉴别手段,若将它们与传统身份认证技术相结合,势必大大提高身份鉴别的安全性和准确性。本文对人脸、掌纹、指纹和虹膜等几种生物识别的基本原理和主要方法进行了综述,重点研究了人脸和掌纹识别的基本理论和相关技术。在基于彩色图像的人脸检测中,提出了在考虑光照影响的情况下,将色度空间与Hough变换相结合的快速人脸检测定位算法。对基于Haar-like特征与AdaBoost学习算法的人脸检测方法进行了探讨,对级联分类器的构建给出了分析和评价,指出了级联分类器构建过程中参数的选择方法。在掌纹识别方面,提出了一种新的基于关键点加权的掌纹识别方法,并在此基础上设计了一个同时利用手的几何特征和掌纹的纹理特征进行掌纹识别的多模态掌纹识别系统。该系统将掌纹识别的过程分为粗匹配和精细匹配两个阶段。在粗匹配阶段对掌纹图像进行预处理,得到手指的几何特征和掌纹的感兴趣区域(ROI),然后利用手指的几何特征进行粗匹配,以筛选出一个较小的候选特征模板集合。在精细匹配阶段,首先利用多通道二维Gabor滤波器对掌纹的ROI区域进行滤波,再对滤波后的图像进行关键点加权,以提取掌纹的局部纹理特征。最后将关键点之间的相对距离作为特征向量,用Mahalanobis距离进行特征匹配,实验验证了多模态掌纹识别系统的识别率以及关键点加权方法的有效性。最后,根据嵌入式应用系统工程化设计的思想,给出了基于嵌入式人像识别门禁、人像考勤机和网络摄像机等应用系统的硬件设计架构和关键环节,分析并总结了嵌入式应用系统设计过程中的一些关键问题及其解决方法。更多还原

【Abstract】 One of remarkable characteristics in twenty-one century is that the personal identification information becomes digital and recessive. It has been a pivotal problem in the computer network and informatics age on how to identity a person exactly and ensure the safety of information. Those traditional approaches of personal identification such as certificates, intellective cards, pass words and keys could not meet the modern society development because of their hidden trouble safety such as loss, forgery, counterfeit and embezzlement etc. The biometrics identification technology can provide a solution. It combines the computer technology with biometric technology, which uses human inherent biological characteristics such as palm-print, face and iris, and behavioral characteristics such as gait, signature and speech to judge a person’s identity in order to replace or strengthen those traditional personal identification approaches.This paper gives a summery dissertation of basic principle and primary method on several biometric identification technologies such as finger-print, iris, palm-print and face. The evaluate standard is also discussed for a biometric identification system. The basic theory and correlative technology are primarily studied about face and palm-print identification. Finally, this paper gives the implement and its important technology in some embedded application system on face recognition.The finger-print recognition is the earliest technology to be studied and used. It can identify a person by the global features and local features on his finger. The global features are those features which can be observed directly including finger-print shape, pattern area, core point, delta point and ridge count. They describe the global finger-print structure. On the other hand, the local features are those features of points including ending points, bifurcation points, ridge divergence points, dots or islands, enclosure points and short ridges. There are several algorithms for finger-print identification. For instance, the point-mode match algorithm, texture mode match algorithm, mixed feature match algorithm, finger image match algorithm etc. The finger-print identification technology is relatively successful, and its veracity is only lower than iris identification. At the same time, the finger-print identification needs person’s cooperation. It is osculated and attacked.The veracity of iris identification is the highest among all biometric identification technologies these days. It is one of the most developing biometric identification technologies. However, it is in limit application area for these reasons such as the costly capturing device, complicated capturing method, feature extraction, feature expression, matching etc. We should do more research on iris image capture under natural eye’s open, taking the masking area of eyelid and process of eyelashes disturbance.It is accepted that the face is the most prevalent and convenient biometric feature used for personal identification. It is non-osculated and non-attacked personal identification method. The cost of face image capturing is very cheap as it can be captured by common lens. Generally, we also judge someone by his or her face. Therefore, face recognition is the most direct and acceptable biometric identification technology.This paper proposes a new algorithm during face detection for color image on condition at the effect of light. It is a fast algorithm which combines the color space with Hough transform for the detection and location of face. The face image will be repaired for light firstly after being captured. And then establishes the ellipse model in YC_bCr_p space based on the skin-color on face. The operation of inflate and erode is completed before the initialized face detection. The region of face can be located rapidly and accurately after the center of pupil is detected by the automatically adapted illegibility Hough transform. The algorithm is validated on the Georgia Tech Face Database.The algorithm for face detection, based on Haar-like feature and Adaboost learning, is also discussed in the paper. Analyses and evaluates the construction of multi-classification. A method is proposed for the selection of parameters during the multi-classification designing. Finally, an embedded application based on AdaBoost and PCA is implemented so as to estimate the performance of the algorithm under the embedded environment.This paper presents a novel approach for personal identification using weighting relative distance of key point scheme on hand images. At the same time, a new multimodal palm-print identification system is given, in which, both hand geometrical features and region of interest (ROI) features are employed and a coarse-to-fine dynamic identification strategy is adopted to implement a reliable and real time personal identification system. The process of palm-print identification is divided into two level stages including coarse level matching and fine level matching. The hand image is preprocessed during coarse level stage. As a result, five hand geometrical features and the ROI are extracted after the coarse level stage. The size of ROI will be transformed into 128*128 pixels before fine level stage. Then, the five hand geometrical features are used to guide the selection of a small set of similar candidate samples at the coarse level matching stage. In the fine level matching stage, we use mutlti-channel 2-D Gabor filters to process the ROI and get some filtered image. And then extract the key points to analyze the local feature structure of palm-print texture information. The key points are computed from sub-images, in size of 32*32 pixels, which are from filtered palm-print image division. The weighting key point approach is used to set value for key point based on the difference of channel. At decision stage, the Mahalanobis distance matching mechanism is employed to match the test sample among the candidate set samples and output the identification result.In the system-design phase we performed several verification tests to find the optimum values for the system parameters. First of all, we use non-WKP approach to test the difference judgment among those channels of multi-channel 2-D filter. And then select those channels whose judgment performance is better. A bigger value is set for the better channel, and a smaller value is set for the worse channel.We compare our approach with non-WKP, WKP, multimodal system based on WKP under the same amount of samples. The experiment result shows that the accurate rate of WKP is better than that of non-WKP. The identification performance of multimodal system is best among the three approaches and its accuracy is 96.41 percent. However, the high accuracy is based on longer time of identification because of its coarse-to-fine processing. We aim to find a new validate approach on palm-print identification in this paper. The research on speed of algorithm and accuracy will be explored for the future.Several embedded application system architectures, based on biometric identification, are discussed in chapter five of this paper. We propose the thought and approach for every embedded application system. At last, we give the process to design a few embedded application system such as the embedded face recognition door manager, embedded face recognition attendance system and network-vidicon.We provide a architect for the single-door and double-access face recognition system. The principle of hardware is proposed for the door manager terminal. The DSP, which is the center of hardware system, is described in detail on how to joint with other components and image capturing module. The work flow of door manager system is also provided in this paper. At the same time, we explain the flow chart for the software algorithm and evaluate the performance of the door manager system. The solutions for real time, safety and explanation are all put forward.We give the principle and method of designing the embedded attendance system based on face recognition door manager. In this section, we give the architect of embedded attendance system and the detailed circuit diagram for those important components such as image sensor and network interface. The work flow of attendance, on how to work with face recognition, is also given in this paper. We give the running parameters of the attendance in the end.A DSP-based network-vidicon design method is given at the end of this paper. In this section, we also design its architecture and the diagram of those important components such as DSP-based board, image capturing module and network interface etc. We illuminate something advanced in our network-vidicon by comparison with those existed products. Finally, some greatly important questions are discussed in the paper.更多还原