Shigefumi Yamada

Irreversible fingerprint template using Minutiae Relation Code with Bloom Filter

Template protected fingerprint authentication techniques have been proposed, which enables to create an irreversible fingerprint template. In this paper, we propose a new irreversible template creation technique using Minutiae Relation Code(MRC) which can describe the minutiae information efficiently, and Bloom Filter which can realize the irreversibility feature. We evaluate the authentication accuracy and security factors such as Shannon Entropy and the number of attack possibilities using FVC2002 and FVC2004 DBs. As a result, our proposed method can achieve 1.8% EER in FVC2002 DB2 with 249 attack possibilities.

Privacy-Preserving Fuzzy Commitment for Biometrics via Layered Error-Correcting Codes

With the widespread development of biometrics, concerns about security and privacy are increasing. In biometrics, template protection technology aims to protect the confidentiality of biometric templates (i.e., enrolled biometric data) by certain conversion. The fuzzy commitment scheme gives a practical way to protect biometric templates using a conventional error-correcting code. The scheme has both concealing and binding of templates, but it has some privacy problems. Specifically, in case of successful matching, stored biometric templates can be revealed. To address such problems, we improve the scheme. Our improvement is to coat with two error-correcting codes. In particular, our scheme can conceal stored biometric templates even in successful matching. Our improved scheme requires just conventional error-correcting codes as in the original scheme, and hence it gives a practical solution for both template security and privacy of biometric templates.

Recovering Attacks Against Linear Sketch in Fuzzy Signature Schemes of ACNS 2015 and 2016

In biometrics, template protection aims to protect the confidentiality of templates (i.e., enrolled biometric data) by certain conversion. At ACNS 2015, as a new approach of template protection, Takahashi et al. proposed a new concept of digital signature, called “fuzzy signature”, that uses biometric data as a private key for securely generating a signature. After that, at ACNS 2016, Matsuda et al. modified the original scheme with several relaxing requirements. A main ingredient of fuzzy signature is “linear sketch”, which incorporates a kind of linear encoding and error correction process to securely output only the difference of signing keys without revealing any biometric data. In this paper, we give recovering attacks against the linear sketch schemes proposed at ACNS 2015 and 2016. Specifically, given encoded data by linear sketch (called a “sketch”), our attacks can directly recover both the signing key and the biometric data embedded in the sketch. Our attacks make use of the special structure that a sketch has the form of a sum of an integral part and a decimal part, and biometric data is embedded in the decimal part. On the other hand, we give a simple countermeasure against our attacks and discuss the effect in both theory and practice.

A Novel Local Feature for Eye Movement Authentication

Eye movement authentication technology has been proposed as a biometric modality, which enables to authenticate a user continuously, and has the counterfeit feature because of the difficulty of the imitation. By using the eye movement authentication, it is possible to realize an automatic authentication system as long as he/she is looking at a display to operate the device. However, the authentication accuracy is still low compared to the other traditional biometric modalities, such as fingerprint, face, and iris. In this paper, we propose the novel local eye movement feature to represent local differences of the captured time-series gazing position data, and we show the proposed feature can work as a complement feature against Mel Frequency Cepstrum Coefficients(MFCC), which is based on the local phase information of eye movement data.We show the classification rate improves from 61% to 82% in the BioEye2015 dataset by using our proposed method on the best case.