基于Gabor小波和NBP算法的手掌静脉识别

Palm Vein Recognition Based on Gabor Wavelet and NBP Algorithm

为了提取手掌静脉图像的纹理特征,并有效提高其识别率,在联合Gabor小波和近邻二值模式(NBP)的基础上提出了一种纹理特征提取方法。该方法利用静脉结构中血管粗细与延伸方向不同的特点,将掌静脉图像感兴趣区域与4尺度、4方向的Gabor小波卷积获得多个幅值特征,并在4个不同的尺度下分别求取均值,获得Gabor尺度均值模式(GSP),在每个GSP分块上使用NBP描述算子来提取局部邻域关系模式(GSPNBP)。然后将这些多尺度、多方向的GSPNBP分块区域的编码序列的总和作为掌静脉特征向量。最后通过求特征向量间汉明距离衡量静脉图像的相似程度来计算识别率,并在PolyU图库和自建图库中进行实验。实验结果显示,该算法获得的识别率最高可分别可达99.7935%和99.3965%,识别时间都在1s以内,有效增强了算法稳健性。

In order to extract the texture features of palm vein image and improve the recognition rate effectively, a texture feature extraction method based on joint Gabor wavelet and neighbor binary pattern （NBP） is proposed. Considering the difference of vein thickness and extension direction in venous structure, this method obtains multiple Gabor-magnitude features by convoluting the region of interest of the palm vein image with Gabor wavelet of four scales and four directions. And the Gabor scale-mean pattern （GSP） is obtained by averaging four different scales. The GSP neighbor binary pattern （GSPNBP） is extracted from each GSP block using the NBP description operator. Then the concatenation of the coding sequences of these multi-scale and multi-direction GSPNBP regions is used as the feature vector of the palm vein. Finally, the similarity of the two vein images is calculated by Hamming distance of the feature vectors, and the experiments are carried out in the PolyU and self-built database, respectively. Experimental results show that the highest recognition rate of this algorithm can be reached to 99. 7935~~ and 99. 3965%, respectively, and the recognition time is less than ls, which effectively enhances the robustness of the algorithm.