Fingerprint Image Preprocessing Technique and Feature Extraction Algorithm

Yang Xiao-Dong,Ning Xin-Bao,Yin Yi-Long

DOI: https://doi.org/10.3321/j.issn:0469-5097.2006.04.004

2006-01-01

Abstract:The biologic identification technology that is delegated by fingerprint has widely developed in recent years.It is recognized to bring a revolution in identification area.The study of fingerprint has high academic vague as well as social benefit,and is becoming one of the research hotspots in many countries.Of the whole Automatic Fingerprint Identification System(AFIS),fingerprint image preprocessing is the core and key step.The input fingerprint image should be preprocessed first before feature extraction and matching can be done afterwards.These processes can influence the speed and accuracy of the entire system. In this paper,we made a deep research on these preprocessing techniques,which included the separation of foreground and background regions,the orientation information extraction,ridge extraction,image enhancement,binarization,image thinning,as well as feature extraction.In these process modules,we brought forward some new algorithms and novel thoughts.Therefore,the AFIS was complemented and perfected in several aspects.According to the characteristic structure that the fingerprint itself is a curving plane,we introduced a separating method using self-adaptive local threshold,which could solve the problems that fingerprint images in areas of different depth could all be separated from background effectively.We also presented an improved non-thorough thinning-image-based feature extraction method.In the premise of not dealing with the ridge anywhere,we extracted the feature points' set of the original fingerprints directly from the thinned images.Then,we repaired these feature points by deleting those fake minutiae brought by some sorts of yawp.This algorithm greatly increased the feature extraction speed.The accurate rate was achieved to above eighty seven percent.But when the fingerprint images were of very poor quality,the inherent flaw of this algorithm would delete some true minutiae.In addition,we also introduced a novel feature extraction method,which was based on the following fingerprint ridgelines.The minutiae were directly extracted from the original gray scale fingerprints that had not been preprocessed,such as binarization and thinning etc.Compared with traditional methods,this one could markedly reduce the accumulated errors produced by preprocessing processes.So,the speed was increased and the accurate rate was enhanced to some degree too.In the end of this paper,we tested the performance of this AFIS under NJU-2000 database.The testing items included seven aspects such as Average Enroll Time,Average Match Time,False Rejection Rate,and False Acceptance Rate,etc.Using the FRR as the function of FAR,we drew the Receiver Operating Curve then.The experiment results showed that this system could achieve better results and performance indexes.The preprocessing time by using these improved methods mentioned above was less than that of the traditional algorithms.After testing many fingerprint images of different qualities,we arrive at the conclusions that the FRR is less than two percent,the FAR is less than zero point zero one percent,and the identification time is not more than two seconds.These performance indexes all reach an advanced level.