手指静脉识别方法研究

发布时间：2019-03-07 20:35

【摘要】：手指静脉识别是一种新兴的基于人体生理特点的生物特征识别技术。该识别技术利用手指掌侧浅静脉进行身份识别。与其他生物特征识别一样,手指静脉识别有四个处理步骤,即图像采集、预处理、特征提取及匹配。尽管手指静脉识别的相关研究取得了一定的进展,但每个处理步骤中仍存在一些问题尚待完善。比如,由于图像采集过程中手指的随机放置,数据库中存在部分倾斜图像,但此问题在现有的图像预处理方法中没有得到足够的重视。同时,现有感兴趣区域提取方法大都是针对来自某一种设备采集的图像设计,当处理来自其他设备采集的图像时,其分割性能将大打折扣。再如,在基于静脉纹路的识别方法中,不完整的纹路和不鲁棒的匹配方法使得识别性能仍有待提高。使用软特征增强主特征区分性是生物特征识别领域中提高识别性能的一种有效方法,但在手指静脉识别中该问题并没有得到应有的关注。另外,由于个体差异和采集设备性能不够理想,手指静脉图像质量存在较大差异,因此如何正确地评价图像质量也是手指静脉识别中需要重视的问题。本文针对手指静脉图像感兴趣区域提取、纹路特征提取、软特征提取、图像质量评价等问题开展了深入研究,主要的工作和贡献如下：(1)研究了一种基于滑动窗口的手指静脉感兴趣区域提取方法。针对手指倾斜问题,提出了一种基于手指中线的倾斜校正方法。在图像校正的基础上,设计了一种基于滑动窗口的方法检测手指关节的位置,并基于此位置确定感兴趣区域的高度。进一步,由手指左右两侧边界的内切边确定感兴趣区域的宽度,至此就可以得到校正的感兴趣区域。(2)提出了一种基于超像素分割的设备无关手指静脉感兴趣区域提取方法。多个不同设备采集的图像在尺寸、背景灰度及噪声的位置、面积、形状等方面存在差异,这些差异使得现有的基于单设备的感兴趣提取方法的性能大打折扣。但多设备采集的图像存在以下两方面的共性：(1)噪声位于手指区域之外；(2)手指边界处背景与手指区域灰度差异较大。同时,超像素分割能够将具有相似灰度值的相邻像素划分到一个块中,这样不仅噪声被隔离,背景和手指区域也能够被划分到不同的块中。因此,提出了一种基于超像素分割的感兴趣区域提取方法,即从超像素边界中追踪出手指边界,从而获得感兴趣区域。(3)研究了一种基于解剖学结构分析的手指静脉识别方法。现有基于静脉纹路的识别方法性能不够理想的主要原因之一是没有对手指静脉的解剖学结构特点和成像特点进行深入的分析和利用。该工作利用静脉纹路具有谷形或半谷形横切面的成像特点,并结合静脉纹路在解剖学上的方向性、宽度变化、连续性等特点,提出了一种基于解剖学结构分析的纹路提取方法。在匹配阶段,为克服手指随机放置造成的大尺度图像平移问题,提出了基于静脉主干的图像校准方法。同时,为了克服小尺度纹路变形问题,提出了一种弹性匹配方法。为了全面地表征静脉纹路,将图像校准过程中的静脉主干重合度和静脉网络的弹性匹配得分进行集成。(4)研究了一种结合软特征的手指静脉识别方法。该工作检测手指远端关节宽度,并将其看作一种软特征。为了融合关节宽度特征和静脉特征,提出了三种框架,即融合框架、过滤框架、混合框架。实验证明,利用手指关节宽度作为软特征能够有效地增强手指静脉特征的识别性能。(5)提出了一种基于支持向量机的手指静脉图像质量评价方法。为了全面的表征图像质量,提出了三种质量特征,即空域梯度特征、对比度特征及信息容量特征。考虑到手指静脉图像质量分类是小样本非线性分类问题,支持向量机被用于该分类问题中。同时,为了克服高、低质量图像间的类别不平衡问题,R-SMOTE技术被用来合成少数的低质量图像。
[Abstract]:Finger vein recognition is a new biological feature recognition technology based on human physiological characteristics. The identification technique uses the finger-palm-side superficial vein for identification. Like other biometric identification, finger vein recognition has four processing steps, image capture, pre-processing, feature extraction, and matching. Although some progress has been made in the research of finger vein recognition, there are still some problems in each processing step. For example, because of the random placement of the finger in the image acquisition process, there is a partial oblique image in the database, but this problem is not given enough attention in the existing image preprocessing method. At the same time, most of the existing region of interest extraction method is for image design from a certain kind of equipment, and when the image acquired from other equipment is processed, its segmentation performance will be greatly reduced. For example, in the vein-based recognition method, incomplete lines and non-robust matching methods make the identification performance still to be improved. The use of soft-feature to enhance the primary feature differentiation is an effective way to improve the recognition performance in the field of biometric identification, but the problem does not receive due attention in finger vein recognition. In addition, because that individual difference and the performance of the acquisition device are not ideal, there is a great difference in the quality of the vein image of the finger, so how to correctly evaluate the image quality is also a problem that needs to be paid attention to in the finger vein recognition. In this paper, a deep study is carried out on the problems of region extraction, texture feature extraction, soft feature extraction, image quality evaluation, etc. of finger vein image, and the main work and contribution are as follows: (1) A method for extracting a region of interest of a finger vein based on a sliding window is studied. In view of the problem of finger tilt, a method of tilt correction based on the mid-line of the finger is proposed. On the basis of image correction, a method for detecting the position of a finger joint based on a sliding window is designed, and the height of the region of interest is determined based on the position. Further, the width of the region of interest is determined by the internal trimming of the boundary between the left and right sides of the finger, so that a corrected region of interest can be obtained. And (2) a method for extracting a device-independent finger vein region of interest based on a super-pixel division is provided. The images acquired by a plurality of different devices are different in size, background gray level and noise location, area, shape, and the like, and the differences enable the existing single-device-based interest extraction method to be greatly reduced in performance. However, there are two aspects: (1) the noise is outside of the finger area; (2) the background of the finger boundary is larger than that of the finger area. At the same time, the super-pixel division can divide the adjacent pixels with similar gradation values into one block, so that not only the noise is isolated, but the background and the finger area can be divided into different blocks. Accordingly, a method for extracting a region of interest based on a super-pixel division is proposed, that is, a finger boundary is tracked from the super-pixel boundary to obtain a region of interest. (3) A method of finger vein recognition based on anatomical structure analysis was studied. One of the main causes of the existing recognition method based on the vein line is that the anatomical structure and the imaging characteristics of the finger vein are not analyzed and utilized in-depth. This work uses vein lines to have the features of the valley-shaped or half-valley cross-section, and combines the characteristics of the vein lines in the anatomy, such as the orientation, the width, the continuity and so on, and puts forward a pattern extraction method based on the analysis of the anatomical structure. In the matching stage, in order to solve the problem of large-scale image translation caused by random placement of fingers, an image calibration method based on a vein trunk is proposed. At the same time, in order to overcome the problem of small-scale grain deformation, an elastic matching method is proposed. In order to characterize the vein pattern comprehensively, the degree of the vein trunk and the elastic matching score of the vein network in the image calibration process are integrated. (4) A method of finger vein recognition combined with soft features was studied. This work detects the distal joint width of the finger and regards it as a soft feature. In ord to blend that width and vein characteristics of the joint, three framework, i. e., a fusion frame, a filter frame, and a mixing frame, are proposed. The experimental results show that the finger joint width is used as a soft feature to effectively enhance the recognition performance of finger vein features. And (5) a method for evaluating the quality of a finger vein image based on a support vector machine is proposed. In order to comprehensively characterize the image quality, three quality characteristics, namely, the spatial gradient characteristics, the contrast characteristics and the information capacity characteristics, are proposed. In view of that problem that the classification of the quality of the vein image of the finger is a non-linear classification of a small sample, a support vector machine is used in the classification problem. At the same time, in order to overcome the problem of the class imbalance between high and low quality images, the R-SMOTE technique is used to synthesize a small number of low quality images.
【学位授予单位】：山东大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】