分辨率LBP的掌纹特征提取  

Palmprint feature extraction based on multiresolution LBP

作　　者：李倩颖[1] 阮秋琦[1] 

机构地区：[1]北京交通大学计算机学院,北京100044

出　　处：《智能系统学报》2010年第6期482-486,共5页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(60973060);高等学校博士学科点专项科研基金资助项目(20080004001)

摘　　要：为了研究NSCT变换(nonsubsampled contourlet transform,NSCT)和局部二值模式(local binary pattern,LBP)在掌纹识别应用方面的可行性和性能,提出了一种基于NSCT变换与局部二值模式相结合的掌纹特征提取算法.该方法能够较好地提取皱褶、乳突纹等掌纹的细节特征,并且能够有效减少掌纹识别中由于图像的平移、旋转和光照对识别结果造成的影响.使用NSCT变换可以稀疏地表示二维奇异曲线并且变换本身具有平移不变性;而LBP算子是一种有效的纹理描述算子并且该算子具有很好的灰度和旋转不变性.首先对掌纹图像进行NSCT变换,然后对变换后的掌纹子图提取局部多分辨率和多尺度的LBP特征.实验结果表明,该算法能够更好地表达掌纹纹线的细节和结构特征,对掌纹图像有更高的鉴别性.In order to study the feasibility and performance of the nonsubsampled contourlet（NSCT） and local binary pattern（LBP）operation in palm recognition,this paper presented a palmprint identification approach,which was the combinations of NSCT and LBP operation.It can effectively extract detail features of palmprint and reduce the influence on the shift,rotation and illumination of palmprint identification.The NSCT provides sparse representation at both spatial and directional resolution and is a fully shift-invariant.LBP is a powerful texture descriptor that is gray-scale and rotation invariant.In this paper,we extracted multiresolution and directional local binary pattern histogram to represent the local features of a palmprint image after NSCT.The results of experiments illustrate that this identification approach is efficient and effectively extract details and structure of palmprint.

关 键 词：掌纹识别 NSCT变换 LBP 平移不变性 旋转不变性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]