稀疏表示的手掌图像识别研究  

Research on Sparse Representation Palmprint Identification

作　　者：翟林[1] 潘新[1] 刘霞[1] 罗小玲[1] 

机构地区：[1]内蒙古农业大学计算机与信息工程学院,内蒙古呼和浩特010018

出　　处：《计算机仿真》2014年第12期334-338,共5页Computer Simulation

基　　金：内蒙古自然科学基金项目(2012MS0919;2012MS0927);中国博士后科学基金项目(.20100480370;201104179);内蒙古高等学校研究基金(NJZY13074);内蒙古农业大学基金项目(JCYJ201201;ND-PYTD 210-9)

摘　　要：针对手掌位置、光照、采集设备等外界因素会影响掌纹图像的识别率以及传统稀疏重构的分类方法计算复杂度高的问题。提出融合双向二维主成分分析((2D)2PCA)与压缩感知的掌纹识别方法,将L1范数最小化重构算法替换成分类正交匹配追踪(COMP)算法,以降低复杂度。首先利用双向二维主成分分析对掌纹图像行列两个方向进行降维,提取特征矩阵,做为压缩感知算法的过完备字典。然后通过分类正交匹配追踪算法(COMP)求解图像在过完备字典上的稀疏表示,以得到一组最优稀疏系数重构每个图像。最后求得测试图像与各类重构图像的最小残差得出分类结果。基于北京交通大学掌纹库的实验结果表明,主成分分析与压缩感知方法可有效降低计算复杂度,对于光照不均匀和有位置变化的掌纹具有一定的鲁棒性,具有良好的掌纹识别性能,可以得到较高的掌纹识别率。Because the recognition can be interfered by palm position,lighting, collection equipment and the sur- rounding complex environment and due to the high computational complexity of traditional sparse reconstruction clas- sification, the paper presented a method which fuses bi - directional two - dimensional principal component analysis （（2D） 2pCA） with compressive sensing palmprint identification. First, the dimensionality of palmprint ranks was re- duced through bi - directional two - dimensional principal component analysis, and the feature matrix was extracted as the overcomplete dictionary of compressive sensing algorithm. Then, the sparse representation of the overcomplete dic- tionary was solved by classification orthogonal matching pursuit algorithm （COMP） to obtain a set of optimal sparse coefficients to reconstruct each image. And finally, the classification result can be gained by comparing the test ima- ges with reconstructed images. Based on the experiments of Beijing Jiaotong University palmprint database, the results show that this method,which reduces the computational complexity and gets a higher palmprint recognition rate, can provide a strong robustness for uniform illumination and palms position change.

关 键 词：掌纹识别 双向二维主成分分析 压缩感知 分类正交匹配追踪算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]