【作者】 李强；

【导师】 杨坤涛；

【作者基本信息】 华中科技大学 ， 光电信息工程， 2010， 博士

【摘要】 静脉身份识别技术是将皮肤下的静脉血管作为身份特征进行身份识别的技术。由于人体静脉血管隐藏于皮肤之下,不易伪造,因此静脉特征识别技术是一种安全性很高的身份识别技术。国内外静脉身份识别大部分集中在手背静脉特征和手指静脉特征,这主要是由于手背及手指静脉图像采集相对容易。由于人体手掌部分静脉血管较多,特征较丰富,更适合作为身份识别特征,目前国内关于此技术的研究还未见报道,为此,本文对基于手掌静脉特征的身份识别技术进行研究,主要包括图像采集、图像预处理、模式识别,具体内容如下：(1)在图像采集部分,为了获取静脉特征,分别搭建被动式热红外成像系统和主动式近红外成像系统,利用热红外成像技术可以获取手背静脉图像。利用主动式近红外图像系统,采用850nmLED作为光源,使用CMOS图像传感器,在传感器之前,增加一片中心波长是850nm的滤光片,可以有效采集手掌、手背、手指、手腕部分的静脉图像。研究了滤光片对消除杂散光影响的作用,研究了LED光强对图像亮度和对比度的的影响,得到合理的LED光强；(2)为了缩短手掌放置位置与图像采集装置的距离,同时获取的图像范围尽可能大,采用广角镜头进行采集。对广角镜头采集的图像产生的几何畸变,提出了一种新型校正方法,与传统方法相比,该方法可以自动生成控制点,提高工作效率,然后利用BP神经网络拟合几何畸变模型。在学习策略方面,本文采用附加动量算法进行学习,与常用的最速梯度下降法相比,克服了其收敛速度缓慢、易陷入局部极小等缺陷;(3)图像预处理阶段包括图像有效区域获取、图像归一化、图像对比度增强、图像去噪等四部分工作。分别采用对比度线性拉伸、直方图均衡化等方法进行增强,然后将CLAHE算法应用到静脉图像,CLAHE方法可以有效的进行自适应对比度增强。在图像去噪阶段,由于静脉具有“曲线”特征,而曲波变换能够很好的表达图像中的“曲线”特征,因此本文提出了一种基于曲波变换去噪算法,与其他图像去噪方法相比,该方法能有效消除噪声：(4)在身份识别部分,本文共采用四种算法：PCA方法、小波能量算法、曲波能量算法、曲波变换和PCA相结合算法。在识别模式下,PCA方法的识别率达到99.4%,但是PCA方法不适于进行身份验证。小波变换可以获得图像在不同分辨率、不同方向下的分布情况,利用Haar小波变换后的能量作为静脉图像的特征进行身份识别和验证,正确识别率(CRR)达到98.8%,EER达到2.6%。与小波变换相比,曲波变换是近几年提出的一种新的多分辨率分析工具,具有很强的的曲线表达能力。实验中利用静脉图像的曲波变换后的能量作为特征,正确识别率(CRR)达到99%,EER达到2.1%,然后研究将曲波变换系数与PCA降维相结合,识别率达到99.6%。本论文从手掌静脉图像的采集、预处理、特征提取与识别三个方面进行研究,在每个阶段,分别提出多种解决办法进行处理,并比较处理结果,研究出了一套有效的手掌静脉身份识别实验装置和算法。更多还原

【Abstract】 Hand vein recognition is the technology that the vast network of blood vessels is taken as the pattern for recognition. The vein pattern is hard to fake as the vein is under a person’s skin. Therefore, the vein recognition is a potentially good biometric. In recently years, most hand vein recognition researchers focus on the vein of the back of hand and finger vein because the vein image of these parts is easy to capture. The palm vein is suitable for biometric as it contains a lot of blood vessels. However, to our knowledge, there is no domestic institute and company has carried out research on palm vein pattern biometric recognition technology. In this paper, a new personal recognition system using vein patterns in the palm side of the hand is proposed. The research includes image acquisition, image preprocessing, and pattern recognition.(1) Because of no publicly available palm vein pattern database for research community, we design our own near infrared palm vein image acquisition system in order to utilize palm vein pattern for recognition. In this system, we use an array of LEDs which emit the infrared light at a wavelength of 850nm to shine infrared light onto the palm side of the hand. At the same side, an IR CCD camera whose spectral response also peaks at a wavelength of around 850nm is used to obtain the image of the palm vein. To dissipate the effect of visible light, an IR filter is mounted in front of the camera’s lens.(2) A short-focus len is used to acquire the palm vein image of view in wide field. However, there is generally nonlinear geometric distortion to some extent as the imaging of object point in the image plane is compared with the ideal imaging. A method of digital image geometry correction based on neural network and automatic generation of reference points has been proposed in this paper. The experiment results indicate that the reference points can be settled quickly and readily with the algorithm of automated generation of reference points. Moreover, neural network fits the distortion model which reaches a high precision. In a word, the method put forward in the paper possesses commendable capability in correcting the geometry distortion of image.(3) In the image pre-processing stage, the research includes the ROI extraction, image normalization, image contrast enhancement, and image denoising. Linear stretching algorithm, histogram equalization algorithm and CLAHE algorithm are used to enhance the image. Curvelets are very good at representing objects with curve-punctuated smoothness. In this paper, the curvelet transform algorithm is proposed for the imag denoising. This algorithm can effectively eliminatie noise compared to other method.(4) In the recognition section, four algorithms are used in this paper:PCA method, wavelet energy algorithm, curvelet energy algorithm, curvelet transform based on PCA algorithm. In recognition mode, Using PCA method, the recognition rate (CRR) is 99.4%, but PCA is not suitable identification mode. Wavelet energy algorithm can obtain images at different resolutions, different directions of the distribution. We use the energy of Haar wavelet transform of the image as vein pattern for identification and recognition, CRR reached 98.8% and EER is 2.6%. Compared with the wavelet transform, curvelet transform has good representation of curve. We use the energy of curvelet transform of the image as vein pattern for identification and recognition, CRR reached 99% and EER is 2.1%. The curvelet transform based on PCA algorithm is used for recognition, and get CRR of 99.6%.In this thesis, a new personal recognition system using palm vein patterns is proposed. This system is convenient to acquire vein images compare to those base on vein pattern of the back of the hand. Using those recognition algothm we proposed in this paper, the palm vein recognition system has excellent performance.更多还原