【作者】 汤永华；

【导师】 苑玮琦；

【作者基本信息】 沈阳工业大学 ， 测试计量技术及仪器， 2015， 博士

【摘要】 人体的一些生理结构,或称其为生物特征,例如脸形、虹膜纹理、指纹、手形、掌纹等,具有唯一性,并且在相当长的时间内具有不变性,不丢失也不被忘记。近年来,在一些场合人体生物特征越来越多地用来取代容易丢失或者仿造的钥匙、各种身份证件、容易忘记或者被窃取的密码等。生物特征在公共安全和人们的日常生产生活中发挥着越来越重要的作用。多模态生物识别是一种基于信息融合技术的生物特征识别技术,应用“取长补短,优势互补”的思想。多模态生物特征识别技术能够全面、有效的提高生物特征识别系统的整体性能,具有很强的鲁棒性,能够达到单一生物特征识别无法或很难达到的品质。本文以手部多模态生物特征识别作为主要研究内容,以自制的手部多模态图像获取装置为基础,提出了嵌入式手部多模态融合问题的解决方案。本文主要工作如下:(1)针对手部生物特征中的手形、掌纹、手掌静脉等特征图像可以通过图像获取装置一次性就可被获取到的特点,在综合分析掌纹和手掌静脉成像原理的基础上,制作了可以用于验证手部多模态识别算法的基于双CCD的手部多模态图像获取装置。该装置配备包括多个波段的可见光和多个波长的近红外LED多光谱光源,而且各波长LED光源都可以根据需要灵活调节发光强度。双CCD相机可以在多个波段LED下,分别同时获取到掌纹图像和手掌静脉图像。利用该装置可以方便的获取到手部三种模态图像的样本,建立相应的图库,验证相关融合识别方法的有效性。为了能够将手部多模态生物特征识别方式更好的应用于日常生活,开发和制作了基于嵌入式系统的双摄像头手部多模态图像获取和处理系统。利用该系统可以进行低分辨率下的手部多模态图像的获取和处理,为手部多模态生物识别方式的实际应用提供了可能。同时,为了能够有效降低手部多模态图像采集所需要的LED扩展光源的体积,同时保证照射到全手掌光线的均匀性,提出利用有限个高亮度LED构建小体积LED扩展光源,然后通过透镜对小体积LED扩展光源进行光线的二次分配,最终达到光强充足和均匀的目的。本文通过对光线二次分配进行详细的理论分析、光线追踪模拟和实验,取得了比较理想的效果。(2)针对非接触条件下获取的掌纹和手掌静脉图像纹理的特点,提出了基于纹理峰谷结构的掌纹掌脉像素融合方法。首先分别将获取到的包含掌纹的全手图像和包含手掌静脉的全手图像进行方向尺寸归一化,然后在包含掌纹的全手图像中提取ROI图像。由于多模态图像采集时间极短,多模态图像在空间没有移位,因此直接利用掌纹ROI图像坐标提取手掌静脉ROI图像。对获取到的掌纹ROI图像和手掌静脉ROI图像分别进行灰度归一化。根据纹理存在的局部极小值特点和不同方向宽度关系,提取纹理结构。随后根据掌纹和掌脉纹理不同情况下的对应关系,分类进行纹理融合。最后对融合后的图像进行识别。在自制的双CCD手部多模态图像获取装置采集的图库上进行的实验结果表明,融合后的图像的识别效果好于单模态下的掌纹识别效果和掌脉识别效果。(3)针对基于双CCD的手部多模态图像获取装置在非接触条件下获取的掌纹和手掌静脉图像的特点,提出了基于小波变换的掌纹和手掌静脉高低频综合数据融合方法。首先对双CCD获取的掌纹图像和手掌静脉图像分别进行ROI提取和配准等预处理。然后按照预先选择的小波基对掌纹ROI和手掌静脉ROI图像分别进行多层小波分解。随后对掌纹和手掌静脉小波分解后的高频和低频系数按照不同策略进行融合。最后通过小波逆变换得到融合图像。实验结果表明,通过融合后的图像进行的识别效果优于单模态掌纹识别和单模态手掌静脉识别,也优于其它融合方法。(4)针对嵌入式手部多模态图像获取与处理系统存在数据处理能力有限等问题,提出在掌纹掌脉数据融合的基础上融入手形识别,即基于决策融合和数据融合的手部三模态双融合。首先利用手形识别方式的速度优势进行决策融合,即通过合适的双阈值将识别对象划分为三类:身份确认类、身份否决类和身份待定。然后根据识别对象所处的类别,决定是否需要进行基于小波变换的掌纹掌脉数据融合识别。实验结果表明,通过综合利用手部三种模态双融合,在嵌入式系统进行手部身份识别的平均时间大大缩短,识别效果有明显改善。更多还原

【Abstract】 Some physiological structure of the human body, or called biometrics, such as face, iris, fingerprint, hand geometry, palmprint, unique, is invariant in a fairly long period of time. And it is not to be lost and not to be forgotten. In recent years, the biology characteristics of human body has been used to replace easily lost or imitation keys, ID, easily forgotten or stolen password and so on, in some occasions more and more.Adn it has played more and more important role in daily life and production of public security and people’s.Multimodal biometric recognition is a biometric identification technology based on information fusion technology, which is the application of "learn from each other, the advantage is complementary" thought. Multimodal biometric recognition technology can effectively improve the overall performance of the biometric system, which has very strong robustness, and can achieve the quality of single biometric unable or very difficult to achieve. In this paper, taking the hand multimodal biometrics as the main research contents, using self-made hand multimodal image acquisition device as the foundation, proposed the solution of embedded hand multimodal fusion problem. In this paper, the main work is as follows:Firstly, according to the characteristics of biological characteristics of hand in hand shape, palm, palm vein image by image acquisition device characteristics can be one-time can be acquired, based on comprehensive analysis of palmprint and palm vein imaging principle, double CCD hand multi modal image acquisition device is produced, which can be used to validate the pattern recognition of hand multi-mode algorithm. The device is equipped with infrared LED includes a plurality of bands in the visible and multi wavelength multi spectral light source, and the wavelength of the LED light source can flexibly adjust the luminous intensity. Dual CCD camera can be in more than one band of LED, respectively, at the same time access to the palmprint and palm vein image. The device can conveniently obtain the hand three modal images of samples, the establishment of the corresponding Gallery, validates the fusion recognition method. In order to be able to be better hand multimodal biometrics methods used in our daily life, hand dual camera embedded system of multi modal image acquisition and processing system is developmented. Low resolution hand multi modal image can be obtained through the system, and it has the possibility for the practical application of hand multimodal biometric modality. At the same time, in order to effectively reduce hand multimodal image acquisition required LED extended source size, while ensuring the uniformity of irradiation to the whole palm light, this paper proposes the construction of small size LED extended source using finite high brightness LED, and then through the lens extends two distribution of light source on small LED, finally it will make the light intensity be sufficient and uniform objective. Based on the analysis of the two distribution of light, with theoretical ray tracing simulation and experiment, the quite ideal effect is be obtained.Secendly, according to the characteristics of non- contact conditions acquisition palmprint and palm vein image texture, palmprint palm vein fusion method is be proposed. Firstly, the full hand image of palmprint and palm vein will be normalized for size and direction, and then extract ROI images from full hand palmprint image. Since the time of multi modality image acquisition is very short, multi modal image no shift in space, so the palm vein image ROI can be directly extracted according to the ROI palmprint image coordinate. The palmprint ROI images and palm vein image ROI images were normalized to gray. According to the characteristics of different direction and width of local minimum values of texture, the texture structure will be extracted. Then according to the corresponding relationship of palmprint and palm vein texture under different circumstances, texture fusion will be made respectively. Finally, the fused image will be identificated. The experimental results show that the palmprint recognition effect of the fused image recognition effect is better than the single mode and palm vein recognition effect, through the self-made double CCD hand multimodal image acquisition device.Thirdly, according to the characteristics of palmprint and palm vein image from the double CCD hand multimodal images non contact acquiring device, wavelet transform fusion method is proposed based on palmprint and palm vein. ROI of palmprint and palm vein image from double CCD will be extracted and registration pretreated firstly. Then according to the pre-selection wavelet, palmprint ROI and palm vein ROI images were multilevel wavelet decomposition. Then the high frequency and low frequency coefficients of the palmprint and palm vein are fused according to different strategies. Finally the fused image is obtained through inverse wavelet transfer. The experimental results show that, recognition effect through the fused images is better than single modal palmprint recognition and single mode of palm vein recognition, and is better than other fusion methods.Fourthly, for data processing capacity of embedded hand multimodal image acquisition and processing system is limited, join the hand shape recognition based on data fusion on the palm and palm vein. First use of hand shape recognition mode speed advantage, through dual threshold appropriate to identify objects into three categories: identity confirmation classes, classes and unknown identity veto. Then according to the recognition of the objects of the category, to decide whether or not to perform wavelet transform fusion recognition based on palmprint palm vein. The experimental results show that, through the comprehensive utilization of three hand hand identification mode, the average time in embedded system is greatly shortened; the recognition effect is obviously improved.更多还原