基于非下采样Contourlet变换和MB_LBP直方图的掌纹检测  

Palm Print Detection Based on Nonsubsampled Contourlet Transform and MB_LBP Histogram Algorithm

作　　者：戴桂平[1] 林洪彬[2] 

机构地区：[1]苏州市职业大学电子信息工程学院,江苏苏州215104 [2]燕山大学电气工程学院,河北秦皇岛066004

出　　处：《传感技术学报》2014年第10期1387-1393,共7页Chinese Journal of Sensors and Actuators

基　　金：国家自然科学基金项目(60970058);河北省自然科学基金项目(E2012203002);苏州市科技计划项目(SGZ2013135);苏州市职业大学青年基金项目(2013SZDQ02)

摘　　要：结合NSCT分解的多尺度、多方向性、各向异性、平移不变性以及MB_LBP算法的局部特征提取、旋转灰度不变性等优势,提出一种基于非下采样Contourlet变换NSCT(Nonsubsampled Contourlet Transform)和多块分区局部二进制模式MB_LBP(Multi-Block Local Binary Patterns)相结合的掌纹检测方法。该方法首先采用NSCT算法对预处理过的掌纹图像进行多尺度、多方向稀疏分解;其次,利用MB3-LBP8,2算子分别提取低频子图及各高频子图分块子区域的MB_LBP直方图特征向量并级联;最后,采用AdaBoost算法测试(NSCT+MB_LBP)的检测性能。仿真结果表明,与(Contourlet+MB_LBP)、(2D-EMD+ICAⅡ)、(Contourlet+NMF)以及单独的MB_LBP相比,该方法能更有效地提取掌纹特征,检测率更高。A novel method based on NSCT( Nonsubsampled Contourlet Transform) and MB_LBP ( Multi￣Block Local Binary Patterns) was proposed to solve palm print detection. The multi￣scale,multi￣direction,anisotropy,shift invari￣ance of NSCT and local texture property,rotation invariance and gray invariance of MB_LBP were utilized to extract the palm print features. Firstly,the preprocessed palm print image was decomposed by NSCT;and then MB3-LBP8,2 was used to respectively extract MB_LBP histogram eigenvectors of each block region of low and high frequency sub￣bands;Finally,the detection performance of the integrated method ( NSCT+MB_LBP ) was tested on the Hong Kong Polytechnic University palm print database through AdaBoost sorting algorithms. Experimental results show that the proposed method not only extracts the palm print features effectively and accurately,but also achieves higher detection rate in comparison with ( Contourlet+MB_LBP ),( 2D-EMD+ICAⅡ),( Contourlet+NMF ) and the single MB_LBP .

关 键 词：掌纹检测 特征提取 非下采样Contourlet变换(NSCT) 多块分区局部二进制模式(MB_LBP) AdaBoost分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]