法庭科学指纹鉴定几个关键问题的研究进展

Research Progress on Several Key Issues of Forensic Fingerprint Identification

指纹鉴定标准、指纹特征和指纹自动识别系统(automatic fi ngerprint identifi cation system,AFIS)均是指纹鉴定中重要的研究内容。近年来,鉴定结论概率化成为了鉴定标准的发展趋势,指纹特征研究也逐步从二级特征走向三级特征,进入到了更微观的领域。然而随着指纹大库的建立以及三级特征的研究,AFIS的发展面临一些技术瓶颈并亟待加入机器学习等新技术。因此,本文对以上三方面相关研究的最新成果展开综述,在总结指纹鉴定存在的问题的同时,对其未来的发展趋势作出几点展望,力求给指纹的研究提供新视野和新思路。

Fingerprint identification relates to relevant standards,involving specific features,developing technologies and the ever-exerting computer-based fingerprint automatic identification system(AFIS).For fingerprint features,those traditional level-2 ones play their roles in increasing juxtaposition to the level-3 ones with which the systematic basic researches have been already carried out on the related pores about their short-term tissue stability or the trackable others.Besides,the traditional level-2 features have been further subdivided and classified into some rare and more-detailed types.About AFIS,the existing version is facing many technical bottlenecks owing to its expansion of capacity and the recognition of numerous microscopic characteristics.Actually,the AFIS now available in China can only standardize the level-2 features,unable to effectively identify and compare those of level-3’s.Even worse,the continuous enrollment of the fingerprint samples is causing the comparison accuracy of AFIS to decline,resulting in occurrence of the close-yet-nonmatched fingerprints which are indicative of two fingerprints,highly similar yet not homologous,commonly appearing more in the triangle zones.Such fingerprints are potential to cause a certain cognitive risk to identify incomplete fingerprints.Promisingly,a fusion algorithm has been developed about fingerprint’s level-2 and level-3 features,realizing new functions such as the in vivo fingerprint detection.Furthermore,the rapid development of computer technology and establishment of fingerprint databases have made machine learning fulfilled to apply into fingerprint identification in academic and actual practice home and abroad.Usually,the machine learning takes large-scale fingerprint data as models for training and verification through different systems so that a likelihood ratio evaluation model is therewith developed to deliver the probability about fingerprint identity,hence bringing forth the fingerprint identification conclusion from absolute to relative.This paper summarizes the latest achievements related to the above aspects,putting forward the problems and envisioning the prospects for future progress trend regarding to fingerprint identification.