单幅近红外手掌图像掌静脉和掌纹多特征识别

摘要

The traditional palm vein and palmprint fusion recognition needs to obtain the palm vein images and the palm-print images separately.However,the palm vein and palmprint structure information actually exists in the same near infrared palm image.Because the palm vein and palmprint structures stay in the different ranges of pixel values,and their local texture information is also distinct,the palm vein and the palmprint can be separated from a single near infrared palm image and then enhanced individually. Firstly, an improved self-guided filter algorithm is proposed to remove the palmprint structure,and the anti-fuzzy detail enhancement model is designed to enhance the image of the palm vein structure.Then, an improved block enhancement algorithm is proposed to enhance the palmprint structure and remove the palm vein structure. Besides,the Sobel operator unsharp masking algorithm is used to highlight the main line structure of the palmprint.Finally, a recognition algorithm is presented based on the image fusion of palm vein and palmprint.Experiments are implemented on the multispectral palmprint database of HongKong Polytechnic University,and the experimental results show that the fusion recognition rate has reached 99.63%. Compared with the other existing algorithms, the equal error rate of the proposed algorithm is reduced by 0.66% in average,which validates the effectiveness of the proposed algorithm.%传统的掌静脉和掌纹图像融合识别一般需分别采集掌静脉和掌纹两类图像,而单幅近红外手掌图像中实际上同时包含了掌静脉和掌纹结构信息.由于二者局部纹理细节差异较大,且像素值分布范围不同,因此,可以先分离再分别增强处理.首先,提出了改进的引导滤波算法以便去除掌纹结构,并设计了反模糊细节增强模型增强掌静脉结构图像;然后,提出了一种改进的分块增强算法,可以在增强掌纹结构图像的同时滤除掌静脉结构信息,再利用基于Sobel算子的反锐化掩模算法以便突出掌纹主线条结构信息;最后,对单幅近红外手掌图像中获取的掌静脉和掌纹图像进行融合识别.在香港理工大学近红外手掌数据库上进行了实验,结果表明:所提出的算法识别率达到了99.63%,与其他已有算法相比等误率平均降低了0.66%,验证了所提出算法的有效性.