近红外手背静脉识别算法研究

【作者】 郑英杰；

【导师】 顾晓东；

【作者基本信息】 复旦大学 ， 电路与系统， 2010， 硕士

【摘要】 人体静脉识别是一项新的非接触式生物特征识别技术,它在安全保护和身份认证领域都有广泛用途。静脉识别的发展开始于2000年前后,系统开发还不够成熟和完善,但是这些都不妨碍研究人员对其的浓厚兴趣。凭借着人类活体静脉独一无二、安全可靠的特性,杜绝了造假或剽窃的可能性。而且在商业化实践的过程中,我们还发现的其相对于其他生物特征识别技术而言,工作稳定,使用便捷舒适的优点,所有这些都更激发了人们探索和完善这项技术的渴望。和人脸识别、指纹识别等其他生物识别技术相似,人体静脉识别技术在商用系统实现的同样面临着许多有待解决的问题。比如现有的静脉识别研究中的静脉图像受环境光照影响问题、算法计算复杂度过高等,因此并不能很好的应用到实际静脉识别产品的开发中。针对现有的问题和局限,本文在以下方面进行了探索,主要贡献为：1.基于分水岭和豪斯道夫距离的手背静脉识别算法该算法采用了分水岭算法直接提取静脉的单像素级骨架特征,利用静脉骨架的端点和交叉点作为静脉模式的特征,并且使用了改进的豪斯道夫距离来作为静脉特征匹配识别算法,并且在自制静脉图库上验证了该新算法的性能和特点。这种算法避免了传统静脉特征提取算法细化骨架特征时带来毛刺和噪声的问题。2.基于Gabor相位编码的手背静脉识别算法该算法分析了现有基于静脉骨架特征的算法对于表达整个手背静脉特征的局限性,提出了基于Gabor相位编码的识别方法。算法利用二维Gabor滤波器提取了静脉图像的相位信息,然后将编码后的相位信息作为静脉特征,并且利用海明距离作为特征匹配识别算法。在不考虑静脉图像样本几何变化的情况下,相对于现有的静脉骨架特征算法取得了很好的识别效果。3.基于局部Gabor相位信息的手背静脉识别算法在Gabor相位编码算法的基础之上,局部Gabor相位信息算法,将Gabor相位编码的局部特征和全局特征结合起来,采用局部异或算子提取相位编码的局部变化,再利用局部变化特征值的直方图信息作为整个静脉特征,使静脉特征模式更加丰富。并且利用卡方距离作为特征匹配识别的算法。实验表明,本算法对图像的几何变化有很好的鲁棒性,提高了静脉识别系统的性能。更多还原

【Abstract】 Vein pattern recognition for human body is a newly coming touch-less biometrics, which holds great applicable vision on security and identification. The research of vein authentication technology started from around the year thousand and seemed to be very appealing to all the researchers although the developing system is not sophisticated.Compared to other kind of biometric authentication technology, it has lots of advantages.Human vein cannot be forged or stolen since its uniqueness and stability. And the hand vein authentication system is also easy and comfortable for people to use.That is the reason why people are putting much more energy and expectation into the research work to promote this technology.Comparing to the face recognition, fingerprint recognition and other biometrics, vein pattern recognition is also facing many problems in consumer system implementation.As there are still some problems such as environment lighting, algorithm complexity etc, vein recognition is really hard to be used in biometric security system.Based on the problems mentioned above, this article has made researches on the following aspects:1.Palm-dorsal vein Biometrics based on watershed algorithmThis approach is based on watershed algorithm to detect the single pixel level skeleton of the hand vein.Using the ending point and the crossing point of the extracted vein skeleton as the feature point, then measure the similarity between the registered vein image and the sampled vein image using Hausdorff distance. Experiments were conducted on our own palm-dorsal vein image database and the results are satisfactory. This approach avoids the noise problem and glitch problem happened in the thinning stage in traditional vein image recognition.2.Palm-dorsal vein recognition based on Gabor phase encoding methodFirstly, this part of the article discusses the shortage of the traditional vein image recognition based on the vein skeleton feature and then presents the vein image recognition based on Gabor phase encoding approach.The approach uses 2D Gabor filter phase encoding to represent the texture feature of the vein image and uses Hamming distance to evaluate the matching degree.Without considering of the geometric transformation, experimental results on our own palm-dorsal vein image database indicates the vein pattern biometric is potentially a useful biometric.3.Local Gabor phase feature for Palm-dorsal vein recognitionThis part of the article presents the local Gabor phase feature for vein recognition approach based on the Gabor encoding method discussed in last section. The approach combined the local Gabor feature with the global feature by using a local XOR pattern operator to extract the local variance feature of the Gabor phase encoding and using histogram method to represent the global feature.Chi-square distance is applied to verify the efficiency of our method, experiments were carried on our own palm-dorsal vein image database.The experimental results show that the local Gabor phase feature could provide sufficient information for vein recognition and robust to geometric transformation.更多还原