Attack And Security Protection To Biometric Template Data

Biometric identification has played an essential role in authentication due to the unprecedented popularization of smart phones and intelligent terminals.Orthogonal to other authentication technologies,biometric identification can offer a convenient,effective and unique method to authentication.Owing to the above advantages and basing on the fields of fingerprint,face,voiceprint and so on,biometric identification is catching increasing attention.Therefore,it is widely used in financial sectors,judicial process and public security.However,most biometric identification systems directly store the users' original biometric template data without template protection.Templates of target users can be exposed easily in hacked databases.It also raises new concerns about the intrinsic meaning and sensitivity of these biometric template data.Considering about these situations,we have an investigation about the classic and widely used biometric identification schemes.In response to these issues about template,two novel strategies to carry out spoofing attacks on biometric identification system via leaking biometric template using genetic algorithm are proposed.Spoofing attack results demonstrate the high vulnerability of biometric template and the need to protect biometric template.Then,to handle the privacy challenges of privacy-preserving biometric identification,we focus on a specific biometric application: speaker recognition,and propose a new efficient and privacy-preserving scheme.Specifically,the emphases of this paper are as follows:· Firstly,two novel,easy-to-understand,low-cost generation strategies to generate or reconstruct speech voice for spoofing attacks are proposed.Orthogonal to most existing studies and researches,our proposed strategies start from the voiceprint template data which can be exposed in hacked databases.The first strategy only uses leaked voiceprint to generate voice.The second strategy combines leaked voiceprint with adversary's speech.Moreover,experimental results on state-of-the-art and speaker recognition techniques clearly demonstrate that our generated voice can completely imitate users and pass the biometric identification.Therefore,this work has successfully revealed the privacy and vulnerability of voiceprint templates.· Secondly,to cope with the above privacy and security challenges,we propose an efficient and privacy-preserving biometric identification scheme for speaker recognition,called AGREE.With our proposed scheme,the biometric template of users cannot be revealed to Io T devices and the accurate biometric identification result is guaranteed.Meanwhile,the increasing time cost for privacy protection is negligible.The detailed security analysis has proven the security of our scheme.Moreover,extensive results show that our propose schemes is correct,effective whether in terms of computation cost or communication consuming.