基于深度学习的污损指纹识别研究

Fouling and Damaged Fingerprint Recognition Based on Deep Learning

随着社会信息化水平的提高及不稳定因素的增加,人们迫切需要更加可靠的识别技术对身份进行认证。因此,利用生物特征进行鉴定已成为时下热潮。其中的指纹识别更是因其方便性和可靠性受到普遍认同。传统的指纹识别方法基于特征点比对寻求相似性,此种方法特征点寻找容易出错,且随着指纹的模糊、破坏、污损或是其他问题,均会使识别率明显降低。针对这些问题,该文提出基于深度卷积神经网络(CNN)的CBF-FFPF(Central Block Fingerprint and Fuzzy Feature Points Fingerprint)算法对污损指纹图像进行分类识别。CBF-FFPF算法提取指纹中心点分块图像及特征点模糊化图,合并后输入CNN网络,进行指纹深层特征识别。将该算法与基于主成分分析(KPCA),超限学习机(ELM)和k近邻分类器(KNN)的指纹识别算法进行比较,实验结果表明,所提出的CBF-FFPF算法对污损指纹识别有更高的识别率和更好的鲁棒性。

With the development of information technology and the increasing demanding of information security, people are urgently in need of more reliable identification techniques for identity authentication. Therefore, the biometric recognition methods have become a compelling issue. Among the methods, the fingerprint identification technique attracts much interest due to its excellent feasibility and reliability performance. The traditional fingerprint recognition method is based on matching feature points. However, this method needs a long time to find the feature points, and suffering the blur, scaling, damage, and other problems, the recognition rate is decreased seriously. To solve these problems, a fouling and damaged fingerprint recognition algorithm named CBF-FFPF （Central Block Fingerprint and ~zzy Feature Points Fingerprint） is proposed, it is based on Convolution Neural Network （CNN） of deep learning. Combining small sub block fingerprint, which takes the fingerprint core point as the center from the thinned image and fuzzy graph of fingerprint feature points, as original image input to obtain the recognition rate. The recognition rate based on CBF-FFPF is compared with the fingerprint identification algorithm based on Kernel Principal Component Analysis （KPCA）, Extreme Learning Machine （ELM）, and K-Nearest Neighbor （KNN）. Experimental results show that fingerprint recognition algorithm CBF- FFPF has higher recognition rate and better robustness.