基于2DGabor小波与BDPCA的掌纹识别

提出一种解决双向主成分分析(BDPCA)中小样本问题的掌纹识别方法。把掌纹感兴趣区域图像经过2DGabor小波变换后得到的每个图像都作为独立的样本,以增加每一类掌纹的样本数。设计一种基于样本散度矩阵的改进BDPCA算法进行特征提取。采用训练样本的k值矩阵代替训练样本的平均值矩阵,从而获得最优投影矩阵。将2DGabor与改进的BDPCA算法相结合进行掌纹识别。在PolyU掌纹库中的实验结果表明,该方法不仅减少了不同训练样本对识别率的影响,而且能够提高识别率,甚至当每类训练样本数仅为1时,也能得到较高的识别率,有效解决了掌纹识别的小样本问题。

Palmprint Recognition Based on 2DGabor Wavelet and BDPCA

A palmprint recognition method which can solve the small sample size problem of bidirectional Principal Component Analysis（PCA）is presented. The implementation procedure of this method is as follows：Each image is obtained by2 DGabor wavelet transform of palmprint Region of Interest（ROI）image as an independent sample,in order to increase the number of the samples of every kind palmprint. This paper designs an improved algorithm based on samples scatter matrix to extract the palmprint features. This algorithm can obtain the best projection matrix by adopting the k values matrix instead of the average values matrix of training samples. The2 DGabor and the improved BDPCA algorithm are combined to identify every palmprint. Experimental results on the PolyU palmprint database demonstrate that the proposed method not only reduces the influence of different training samples on recognition rate,but also increases the rate,especially it has great performance when the number of training samples is1. The method effectively solves the small sample size problem of palmprint recognition.