【作者】 吴微；

【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， 2013， 博士

【摘要】 人体的手掌静脉具有稳定性、唯一性的特点，是一种可用作身份鉴别的生物特征。掌脉隐藏在表皮下，在可见光下无法拍摄，却可以在近红外光下拍摄，其结构复杂很难被复制。掌脉拍摄需将手掌伸平，而人手在自然状态下处于半握拳状态，因此无法窃取拍摄掌脉图像，非活体手掌因血液停止流动将不能通过认证，所以掌脉又是一种可以作为“活体识别”的生物特征。这些使得掌脉成为一种安全性好的生物特征。作为生物特征识别领域的前沿课题，手掌静脉识别技术以其高安全性等优势拥有着广泛的应用前景，成为近几年的研究热点。目前掌脉识别的研究还处于实验研究阶段，要开发出真正鲁棒、实用的掌脉识别系统，还需要解决许多问题。本文做了以下主要工作：（1）针对识别性能最佳的主动光源波长选择问题，从两个角度对典型的掌脉识别成像波长：760nm、850nm、890nm、940nm进行了比较。从掌脉特征提取角度，建立了基于FDR(Fisher判别率)的掌脉成像清晰度模型，对4种波长拍摄的掌脉清晰度进行比较；从特征匹配角度，以3种典型的生物特征识别算法对4种波长拍摄的掌脉图像进行识别性能比较。模型选择和典型算法的实验结果都表明：850nm优于其他3种波长。（2）针对传统掌脉识别以掌心区作为感兴趣区域(Region Of Interest,ROI)，一直受到部分人群掌心区掌脉成像不清晰的困扰，以致影响整个识别系统的性能的问题，本文研究了掌脉图像感兴趣区域的选择和定位问题。对手掌上的三个区域：掌心区、大鱼际区和小鱼际区进行医学分析和三区域静脉吸光量对比实验，选择出大鱼际区作为ROI区域。进而，提出基于大鱼际区的感兴趣区域定位方法。以靠近手腕侧最大内切圆确定手掌上的两个稳定特征点，利用这两个稳定特征点确定ROI。（3）针对手掌静脉识别系统的安全性问题，提出了一种基于灰度曲面匹配的接触式快速掌脉识别算法。首先对手掌静脉图像提取感兴趣区域，将感兴趣区域等分为若干个子区域，计算每个子区域像素灰度平均值作为该子区域灰度值，以各子区域灰度值构建待匹配图像。匹配时对两个待匹配灰度曲面中的像素灰度做差，得到灰度差曲面，求出该灰度差曲面的方差，将此方差作为衡量两个掌脉特征曲面之间距离的依据，并据此判定两幅掌脉图像是否来自同一只手。实验结果表明：方法具有快速性，为进一步提高掌脉识别系统的安全性提供了一条可行途径。（4）针对非接触采集的约束性降低可能会导致图像变形，同类图像类内差别增大，识别效果不佳的问题，提出了一种基于分块和偏最小二乘的非接触式手掌静脉识别方法。采用分块算法对图像进行快速降维，同时部分解决图像旋转平移问题。用偏最小二乘算法提取掌脉图像中灰度值变异大的若干方向，省去灰度值变异小的方向，且所提取的主元与类别信息相关性最大。使样本最大限度克服旋转、平移、比例缩放以及光照改变造成的类间差距，向最有利于分类的子空间投影以得到稳定的特征向量，同时再次降维，最后利用欧氏距离进行分类识别。实验结果表明：与传统掌脉识别方法相比，该方法能有效地提高正确识别率，降低误拒率。更多还原

【Abstract】 Human palm vein is a permanent and unique physiological feature for biometric. Ithides under the skin and its distribution is hard to copy; it can be captured under nearinfrared but can’t be captured under visiable light; palm vein can be captured as peoplehold hand flat while the nature condition of hand is half clenched fists state, which makespalm vein image difficult for candid shooting. Palm vein in broken palm or dead bodycan’t be used for identification for no blood flow. For this reason, palm vein can be usedfor a “liveness detection” physiological feature for biometric. All these characters makepalm vein a high security physiological feature for biometric. As a leading subject ofbiometrics technology, palm vein image recognition has a wide prospect of application forits outstanding characters, such as high security. Palm vein image recognition has alreadybeen a new research focus in recent years.At present, study of palm vein recognition is on the stage of experiment. Manyproblems need tobe solved to develop real rubost and practical system. This paper doessome work as follows:To the question of which band is the best for an active light in the recognition system,this paper does choice in the typical wavelength of palm vein recognition:760nm,850nm,890nm and940nm. This study solves this problem from two angles. From the angle offeature extraction, establish a model of palm vein image definition and do the choice of thefour kinds of wavelength according to the model. From the angle of feature matching,compare the recognition performance of palm vein image with four kinds of wavelengthseparately by three typical biological identification algorithms. The experiment results ofthe model and3algorithms show that850nm is the optimal wavelength.Traditional palm vein recognition systems use the center of palm as region ofinteresting (ROI), which is troubled with fuzziness of palm vein on palm center area.Consequently, the fuzziness influences the performance of the system. This paper doessome study on the selection and location of ROI. It does medical analysis and contrast experiment of light absorption on the three areas of the palm: the center area, the thenararea and the hypothenar area. The thenar area is selected as the main scope of ROI.According to this result, this paper proposes a location method based on the thenar area.The method finds two stable characteristic points and squares the ROI with these twopoints.To the secrity question of contact palm vein recognition system, an algorithm basedon blocking and grayscale surface matching is proposed. The algorithm extracts region ofinteresting (ROI) of palm vein image firstly. Then, the ROI is equally divided into severalsub-regions. The algorithm computes the average value of the grayscale of everysub-region. These average values construct an image for matching. At the stage ofmatching, the algorithm computes the difference of the corresponding pixels from twomatching images and gets the grayscale difference surface. It calculates the variance of thegrayscale difference surface and considers this variance as the distance between twofeature surfaces. At last, it decides whether these two images come from the same hand ornot according to the variance. The experiment results show that the method is fast andproposes a way of high safety.Low restriction of contactless palm vein collection may cause image deformation andintra-class different increase, which consequently influence the performance of the system.This paper proposes an algorithm of contactless palm vein recognition based on blockingand partial least squares. The algorithm of image block is used to rapidly reduce thedimension firstly. At the same time, the algorithm of image block can settle image rotationand translation in some degree. Then, the algorithm of partial least squares is used toextract some directions with sharp grayscale variation and omits some directions withweak grayscale variation. The extractived components have the maximum relationshipwith the classical information. The algorithm makes the samples overcome the imagerotation, translation, scale and illumination change in intra-class in the maximum limit. Itprojects to the subspace which is the most beneficial to classification and gets the stablecharacter vectors. The algorithm of partial least squares reduces dimensions at the sametime. At last, Euclidean distance is used for classification. The experiment results showthat compared with the traditional method, the proposed method increases the CorrectRecognition Rate (CRR) and decreases the false rejection rate (FRR).更多还原