【作者】 张建新；

【导师】 欧宗瑛；

【作者基本信息】 大连理工大学 ， 机械设计及理论， 2009， 博士

【摘要】 基于生物特征识别的身份鉴别技术提供了一种高可靠性、高稳定性的身份鉴别方式。掌纹识别是一种相对较新的生物特征识别技术,但其发展非常迅速,现已成为生物特征识别技术领域中重要一员。研究掌纹识别技术具有重要的理论意义和巨大的应用价值,它涉及计算机视觉,模式识别,图像处理等多门学科,掌纹识别技术的研究有助于这些学科的发展;在信息安全、公共安全、法律等领域,掌纹识别技术均具有潜在的应用前景。目前尽管掌纹识别技术已取得了很大进展,但在实际应用中仍然存在许多问题亟待解决。本文对掌纹识别相关理论和算法进行了深入研究,主要工作和贡献如下:(1)提出了一种基于Log-Gabor小波相位一致的掌纹线特征提取方法。掌纹线特征是掌纹识别中最基本、最直观的特征,但由于掌纹线的特殊性和复杂性,如何有效提取掌纹线特征一直是掌纹识别的难点。本文从频率域角度考虑,使用Log-Gabor小波相位一致方法提取掌纹图像线特征,包括掌纹方向相位一致特征和掌纹整体相位一致特征。该方法提取的线特征可有效包含纹线的结构信息、强度信息和宽度信息,且提取的线特征比较稳定。在掌纹整体相位一致特征图像上进一步检测出具体的掌纹线,仅使用掌纹线结构特征来表示和识别掌纹,实验结果验证了Log-Gabor小波相位一致提取掌纹线特征用于识别的有效性。(2)提出了基于区域方向码的掌纹线特征表示方法。该方法首先将6个方向的掌纹方向相位一致特征图像均匀分块,对于同一区域的六个子块图像,将具有最大灰度平均值的那个子块所对应的特征图像方向作为该区域主方向,然后采用三位的二进制码对该主方向编码。掌纹区域方向码综合描述了纹线的灰度变化强度及其方向性信息,因而能更好地区分掌纹图像;此外,对区域的方向编码使该方法受图像平移和旋转影响也较小。在PolyU掌纹数据库上的实验结果验证了掌纹区域方向码方法的有效性。(3)提出了一种基于散度差判别局部保持投影(Scatter-difference DiscriminantLocality Preserving Projections,SDLPP)的掌纹识别算法。子空间方法具有特征描述性强、计算代价小和良好可分性等优点,已被广泛应用于生物特征识别。线性鉴别分析(LDA)和局部保持投影(LPP)是其中比较经典的线性方法,LDA注重图像数据间的可分性,LPP则更关注于数据的局部关系。本文结合LDA的判别性思想和LPP注重局部邻域关系的思想,提出了散度差判别局部保持投影算法,并对SDLPP中影响图像分类的主要参数进行了研究分析。SDLPP的目的是使同类图像在低维空间中保持紧致的局部关系,同时使那些在图像空间中距离近的非同类图像在低维投影空间中相对远一些,从而使各类图像能获得更好地区分。在公开掌纹数据库上的实验结果表明提出的SDLPP算法较LDA、LPP具有更优的识别性能,对参数的分析则有助于确定SDLPP的最优参数。此外,论文还使用核方法在理论上将SDLPP扩展得到相应非线性算法,称为核散度差判别局部保持投影(Kernel SDLPP,KSDLPP)。(4)设计了一种适合于智能移动设备的掌纹识别算法,并以该算法为核心开发了联想ET980智能手机实时掌纹身份验证系统。算法首先设计了一种适于移动设备的掌纹图像采集方式并给出了相应定位分割方法,这保证了移动设备下掌纹验证系统的可行性;在识别过程中加入光照预处理,一定程度上消除了外部光照变化的影响;采用Gabor小波变换和子空间方法的结合算法来提取掌纹特征,并用遗传算法对Gabor特征节点的数目和位置做优化选取,在保证识别精度同时显著地提高了识别效率。论文算法在联想ET980手机上可获得等错误率为3.42%的识别精度,实现的掌纹验证系统也满足识别系统的实时性要求。更多还原

【Abstract】 Biometric technology provides a highly reliable and robust approach to personal verification. Palmprint recognition technology is a relatively new biometric technology but develops rapidly. Palmprint recognition technology is one of the most active and challenging research fields, and closely related to many disciplines such as Computer Vision, Pattern Recognition and Image Processing etc. Its research achievements would greatly contribute to the development of these disciplines. It is believed that automatic palmprint recognition would have a great deal of potential applications in information security, public security and law enforcement etc. Though palmprint recognition technology has made much progress, many problems still remain unsolved in practical applications. In this dissertation, the theory and algorithms of palmprint recognition are studied. The main contributions of the research work are as follows:(1) A Log-Gabor wavelet phase congruency based palmprint line feature extraction approach is presented. Line feature is the most essential and remarkable feature of palmprint image, but its effective extraction is still a difficulty due to palm-lines’ speciality and complexity. In this dissertation, a frequency domain algorithm called Log-Gabor wavelet phase congruency is adopt to extract palmprint line feature, including palmprint directional phase congruency feature (PDPCF) and palmprint global phase congruency feature(PGPCF). The proposed method can not only simultaneously extract structure, strength and width information of palm-lines, but provide well feature location accuracy. Palm-lines are then detected from PGPCF image and employed to recognition palmprint images. Experiment results show the effectiveness of the proposed line feature extraction approach.(2) A palmprint region-based orientation code (PROC) scheme is proposed for palmprint line feature representation. In PROC, each PDPCF image is evenly divided into small sub-blocks. The orientation of one palmprint sub-region, which can be determined by comparison of the grey statistic results in six directions, is coded using three binary digits. PROC combines palm-lines’ strength feature and orientation feature. It not only can provide better discriminant capability, but also is robust to the translation and rotation of palmprint image. Experiment results on PolyU palmprint database illuminate that the proposed PROC algorithm is effective for palmprint recognition.(3) A scatter-difference discriminant locality preserving projections (SDLPP) algorithm is put forward for palmprint image recognition. Subspace methods have been widely applied for biometric recognition for their computational convenience, good feature representation and discrimination capability. Linear discriminant analysis (LDA) and locality preserving projections (LPP) are two classic linear methods. LDA focus on the separability of data while LPP pays more attention to samples’ local relationship. SDLPP combines the basic ideas of LDA and LPP, aiming at preserving local relationship of image data from the same class while making the image classes nearby in image space be far from each other in projection space. The dissertation also presents a parameter selection method for SDLPP and extends SDLPP into its nonlinear form by the kernel theory. Experiment results illuminate that SDLPP outperforms LDA and LPP on recognition accuracy and the parameter selection method is effective.(4) A mobile-based palmprint recognition algorithm is proposed, and then an embedded palmprint verification system running on Lenovo ET980 intelligent phone is developed. Firstly, the dissertation designs a novel palmprint image acquisition style and corresponding location and segmentation method, which ensures the feasibility of mobile-based palmprint recognition. Secondly, an illumination preprocessing operation is added to reduce the illumination variation influence. Finally, Gabor wavelet and subspace method are combined to extract palmprint feature. Then a genetic algorithm is employed to optimize the number and location of Gabor nodes, which can significantly improve the recognition efficiency. The proposed palmprint algorithm on Lenovo ET980 phone can achieve the recognition accuracy of EER=3.42%. The developed mobile-based palmprint verification system also satisfies the time requirement of a real-time verification system.更多还原