应用二维EMD和独立成分分析的掌纹识别

Approach for palm print recognition based on two-dimensional Empirical Mode Decomposition and ICA

提出一种基于二维经验模式分解(Two-dimensional Empirical Mode Decomposition,2-DEMD)和独立成分分析(Independent Comment Analysis,ICA)相结合的掌纹识别新方法。利用2-DEMD自适应的时频局域化多尺度和ICAII表征数据的高阶统计特性来提取掌纹特征。首先,对预处理过的掌纹图像进行2-DEMD分解得到多层本征模函数(Intrinsic Mode Function,IMF);其次,利用基于PCA(Principal Component Analysis)降维处理的FastICAII算法提取IMF子图像集的掌纹特征基向量;最后,设计实验测试(2-DEMD+ICAII)的识别性能。实验结果表明,该方法能更有效地提取掌纹特征,与传统的ICAII相比,具有重构图像信噪比好、识别率高等优点。

A novel method based on Two-dimensional Empirical Mode Decomposition（2-D EMD） and Independent Comment Analysis（ICA） is proposed to solve palm print recognition.The adaptive time-frequency localization of 2-D EMD and higher-order statistical independency of ICA II are utilized to extract the palm print features.Firstly,the preprocessed palm print image is decomposed into some Intrinsic Mode Functions（IMFs） by 2-D EMD,and then the palm print feature subspaces of IMF subimages matrix are obtained by a fast fixed-point algorithm for ICA II（Fast ICA II）,before which Principal Component Analysis（PCA） is used to decrease the dimensions of input images matrix.Finally,the recognition performance of the integrated method（2-D EMD＋ICA II） is tested on the Hong Kong Polytechnic University palm print database.Experimental results show that,compared with ICA II,the proposed method not only can more effectively and accurately extract the palm print features,but also achieves superior Signal-noise-ratio（SNR）of the reconstructed image and higher recognition rate.