Tetsushi Ohki

Optimal sequential fusion for multibiometric cryptosystems

We propose a general framework for feature level sequential fusion.We propose an optimal algorithm for feature level sequential fusion.The proposed algorithm does not output scores but appropriately sets a threshold.We prove that the proposed algorithm is optimal by showing its equivalence to SPRT.We show the effectiveness of the proposed algorithm through experiments. Biometric cryptosystems have been widely studied in the literature to protect biometric templates. To ensure sufficient security of the biometric cryptosystem against the offline brute-force attack (also called the FAR attack), it is critical to reduce FAR of the system. One of the most effective approaches to improve the accuracy is multibiometric fusion, which can be divided into three categories: feature level fusion, score level fusion, and decision level fusion. Among them, only feature level fusion can be applied to the biometric cryptosystem for security and accuracy reasons. Conventional feature level fusion schemes, however, require a user to input all of the enrolled biometric samples at each time of authentication, and make the system inconvenient.In this paper, we first propose a general framework for feature level sequential fusion, which combines biometric features and makes a decision each time a user inputs a biometric sample. We then propose a feature level sequential fusion algorithm that can minimize the average number of input, and prove its optimality theoretically. We apply the proposed scheme to the fuzzy commitment scheme, and demonstrate its effectiveness through experiments using the finger-vein dataset that contains six fingers from 505 subjects. We also analyze the security of the proposed scheme against various attacks: attacks that exploit the relationship between multiple protected templates, the soft-decoding attack, the statistical attack, and the decodability attack.

Theoretical vulnerability in likelihood-ratio-based biometric verification

Impersonation by impostors is one of the representative security issues on biometric authentication system. A wolf attack is an attack on biometrics system using a wolf that can be falsely accepted as a match with multiple templates. False acceptance rate (FAR) which has been a conventional standard measure to quantify the average error rates of detecting the impersonation has not taken into consideration that impostors could use artefacts instead of templates generated from an individual. The wolf attack probability (WAP) is thus used as a new measure for evaluating the security of biometric authentication. In this paper, we focus on the vulnerability of likelihood-ratio-based biometric verification scheme that is known as optimal similarity measure in terms of average error rates. First, we present theoretical analysis of a likelihood-ratio-based biometric verification system and show the existence of wolf features under the assumption that there is an approximation error between background model and true feature distribution. Second, we propose a new wolf attack scheme that can achieve 60% of WAP. Furthermore, we empirically evaluate the proposed wolf attack using real biometric data from ATR speech database.

A consideration on a common template-based biometric cryptosystem using on-line signatures

Recently, with the rapid spread of smartphones and tablet PCs, biometric authentication using handwritten signatures has been attracting much interest. However, little research has been reported on biometric template protection, which is indispensable for protecting ones privacy, especially when those devices are used in a network environment. Therefore, as a means of (biometric) template protection, we propose a common template-based biometric cryptosystem that is suitable for signature-based biometric authentication. The results of the simulation using signature data confirmed the effectiveness of the proposed method.

Automatic Examination-Based Whitelist Generation for XSS Attack Detection

When faced with cross-site scripting (XSS) attacks, it is difficult to counter all malicious inputs such that they are rendered completely harmless. In such situations, the introduction of a whitelist-based XSS countermeasure is considered to be an effective and robust approach. However, as the behavior of current web applications is complex, it is difficult to theoretically generate the necessary and sufficient whitelists. To this end, we propose an examination-based approach for whitelist generation instead of a theory-based one. We focus on software tests that are always performed during the final stage of the development process and establish a method to automatically generate whitelists that are consistent with the specifications of each web application. By adding the function for whitelist generation on a web application’s test tool, a whitelist can be generated without changing the development process of a conventional web application. We implement our proposed method and evaluate its effectiveness.

A Study on Human Reflex-Based Biometric Authentication Using Eye-Head Coordination

Biometric information can be easily leaked and/or copied. Therefore, the biometric information used for biometric authentication should be kept secure. To cope with this issue, we have proposed a user authentication system using a human reflex response. It is assumed that even if people know someone’s reflex characteristics, it is difficult to impersonate that individual, as anyone cannot control his/her reflexes. In this study, we discuss a biometric authentication system using eye-head coordination as a particular instance of reflex-based authentication. The availability of the proposed authentication system is evaluated through fundamental experiments.

Face/Fingerphoto Spoof Detection under Noisy Conditions by using Deep Convolutional Neural Network.

Most of the generic camera based biometrics systems, such as face recognition systems, are vulnerable to print/photo attacks. Spoof detection, which is to discriminate between live biometric information and attacks, has received increasing attentions recently. However, almost all the previous studies have not concerned the influence of the image distortion caused by the camera defocus or hand movements during image capturing. In this research, we first investigate local texture based anti-spoofing methods including existing popular methods (but changing some of the parameters) by using publicly available spoofed face/finger photo/video databases. Secondly, we investigate the spoof detection under the camera defocus or hand movements during image capturing. To simulate image distortion caused by camera defocus or hand movements, we create blurred test images by applying image filters (Gaussian blur or motion blur filters) to the test datasets. Our experimental results demonstrate that modifications of the existing methods (LBP, LPQ, DCNN) or the parameter tuning can achieve less than 1/10 of HTER(half total error rate)compared to the existing results. Among the investigated methods, the DCNN (AlexNet) can achieve the stable accuracy under the increasing intensity of the blurring noises.

A Secure and Practical Signature Scheme for Blockchain Based on Biometrics

In a blockchain system, a blockchain transaction is protected against forgery by adding a digital signature. By digital signature verification, we can confirm that a creator of a transaction has a correct private key. However, in some critical fields, we need to prove that a creator of a transaction is a proper user. In such a case, the conventional digital signature verification cannot achieve sufficient security. Furthermore, a system that combines blockchain and IoT has been proposed. However, since an IoT device in this system automatically generates a blockchain transaction, reliable creator verification is challenging issue. To achieve reliable creator verification in the IoT blockchain system, we propose a new signature scheme for blockchain. Our contributions are as follows: (1) We propose a new secure and practical signature scheme. (2) We implement our signature scheme for an IoT blockchain system and evaluate the security and the practicality of our scheme.

Theoretical vulnerabilities in map speaker adaptation

We analyze the theoretical vulnerability of maximum a posteriori(MAP) speaker adaptation, which is widely used in practical speaker recognition systems. First, we proved that there exist a set of feature vectors, what are called wolves, which can impersonate almost all the registered speakers with probability asymptotically close to 1 with at most two trials. Second, our experiment shows that the wolves with appropriate parameters achieved 0.99 of successful impersonation rate on Spear speaker recognition toolkit with ATR speech database.

A Study on Autoencoder-based Reconstruction Method for Wi-Fi Location Data with Erasures

Anonymization is one of the major processes to protect location-based services (LBS) from privacy leakage. However, there are many discussions about de-anonymization attacks to LBS and whether anonymization processing is a sufficient countermeasure for privacy leakage. In this paper, we proposed a novel method to reconstruct the location of user considering the time series using the Markov Transition Field (MTF) and Denoising Auto Encoder (DAE). We also focused on Wi-Fi location data including many erasures errors. We conducted an evaluation of de-anonymization attack using our reconstruction method to the Wi-Fi location dataset that was consisted of 10000 devices / four weeks in the four wards of Tokyo. We confirmed that the successful attack rate (SAR) was 24% when the number of candidate devices was 100 and 6\% when that was 10000 devices.