Research On Small Area Fingerprint Recognition Based On Deep Learning And Multi-feature Fusion

With the rapid development of Internet and popularity of smart devices,lightweight and miniaturization has become the trend of fingerprint acquisition,which makes the image size per single acquisition are getting smaller.The classical algorithm of conventional fingerprint recognition has become mature,but for the problem of feature extraction and matching of small area fingerprints,there is no accepted method.Fingerprint recognition is a complex process,mainly includes preprocessing,feature extraction,feature matching and other steps,which means single improvement can't effectively enhance the results.If we apply the traditional method to small area fingerprint directly,it will cause the sharp fall of recognition rate.The main problem of small area fingerprint recognition is that the effective information contained in small area fingerprint is greatly reduced compared with common fingerprint image.Therefore,how to make the most of the information in small area fingerprint image and improve the reliability of extracted feature,is an urgent problem to be solved,which is also the focus of this study.This paper presents a small area fingerprint recognition method based on deep learning and multi feature fusion,we improved the algorithm of multiple parts.First,to deal with the problem of fewer features,we designed an orientation extraction neural network and a minutiae extraction neural network to obtain higher quality orientation and minutiae feature.Second,in order to solve the matching problems of minutiae,which located in the edge of the image,we proposed a minutiae matching operator based on selective extension,which optimized the performance of minutiae matching in the edge area of the image.What's more,we added orientation,frequency,energy,coherence and other features to the calculation of matching score which improved the accuracy and reliability of the identification.Finally,we evaluated the proposed methods through a set of experiments.Experiments show that the proposed method outperforms other common methods in XDFinger small are fingerprint database.What's more,the proposed orientation and minutiae neural network also reached a leading level in database FVC 2004 and NIST SD27.