摘要
针对虹膜识别经验模态分解(EMD)和局部均值分解(LMD)方法具有无法兼顾分解速度和包含小误差的缺点,提出了将分段三次Hermite多项式插值引入局部均值分解(PCHIP-LMD)的虹膜识别方法来提高识别准确率。针对虹膜纹理的分布特性,利用PCHIP-LMD对归一化的虹膜图像逐行分解,得到不同尺度的分量图像;通过提取有效的分量图像将其二值化为特征图像。然后用Hamming距离对特征图像进行移位匹配,得到匹配向量。最后计算匹配向量的改进标准差,用此标准差进行虹膜识别。对CASIA1.0、CASIA2.0、CASIA3.0-Interval、MMU1图像库进行了识别试验,结果显示识别率分别达到了99.968 1%、99.884 5%、99.993 7%、99.878 2%。实验结果表明:该方法消除了虹膜特征提取时的高频噪声,有效提取了图像的二值特征,与EMD和LMD方法相比,识别速度,识别准确率和鲁棒性均有极大提高。
As Empirical Mode Decomposition （EMD） and Local Mean Decomposition （LMD） cannot take care of both the decomposition rate and the smallest error simultaneously, a fast and effective method combined Piecewise Cubic Hermite Interpolating Polynomial with the LMD（PCHIP-LMD） was proposed to improve the precision of iris recognition. According to the distribution characteristics of iris textures, the PCHIP-LMD method was used to decomposed a normalized recognition image line by line to generate the component image with different scales. Then, the feature image of iris was ob- tained by binarization of useful components for the iris recognition. Furthermore, the Hamming dis- tance was used to match the feature image by horizontal and vertical shifts to obtain the matched vec- tors. Finally, the improved standard deviation of the matching vector was calculated and was used to iris recognition. This method was used in CASIA1.0, CASIA2.0, CASIA3.0- Interval and MMU1 database and obtained results show that the correct recognition rates are achieved respectively 99. 968 1%, 99. 8845%, 99. 993 5%, 99. 878 2%. These experimental results demonstrate that the proposed method eliminates the high frequency noise when the iris feature is extracted and obtains the binary feature of the image effectively, which have the advantages of higher speeds, higher recognition rates and better robustness.
出处
《光学精密工程》
EI
CAS
CSCD
北大核心
2013年第1期197-206,共10页
Optics and Precision Engineering
基金
中科院知识创新工程资助项目(No.KGCX2-YW-911-2)
关键词
虹膜识别
经验模态分解
局部均值分解
HAMMING距离
移位匹配
分段三次Hermite多项式插值
iris recognition
Empirical Mode Decomposition （EMD）
Local Mean Decomposition （LMD）
Hamming distance
shifting match
Piecewise Cubic Hermite Interpolating Pol- ynomial（PCHIP）
