Ulrich Scherhag

On the vulnerability of face recognition systems towards morphed face attacks

Morphed face images are artificially generated images, which blend the facial images of two or more different data subjects into one. The resulting morphed image resembles the constituent faces, both in visual and feature representation. If a morphed image is enroled as a probe in a biometric system, the data subjects contributing to the morphed image will be verified against the enroled probe. As a result of this infiltration, which is referred to as morphed face attack, the unambiguous assignment of data subjects is not warranted, i.e. the unique link between subject and probe is annulled. In this work, we investigate the vulnerability of biometric systems to such morphed face attacks by evaluating the techniques proposed to detect morphed face images. We create two new databases by printing and scanning digitally morphed images using two different types of scanners, a flatbed scanner and a line scanner. Further, the newly created databases are employed to study the vulnerability of state-of-the-art face recognition systems with a comprehensive evaluation.

Is your biometric system robust to morphing attaeks

The wide deployment of biometric recognition systems has raised several concerns regarding their security. Among other threats, morphing attacks consist of the infiltration of artificial images created using biometric information of two or more subjects. These morphed images are hence positively matched to several subjects. Recent studies have shown that such images pose a concrete threat to civil security: wanted criminal offenders can use an authentic passport to enter a country with a false identity. However, there is still no quantitative manner to analyse this threat. We address this shortcoming by proposing a new framework for the evaluation of the vulnerability of biometric systems to morphing attacks. The experimental analysis on real systems based on face, iris and fingerprint shows that even systems providing high verification accuracy are vulnerable to this kind of attacks, depending on the verification threshold and the shape of the mated and non-mated score distributions.

Biometric Systems under Morphing Attacks: Assessment of Morphing Techniques and Vulnerability Reporting

With the widespread deployment of biometric recognition systems, the interest in attacking these systems is increasing. One of the easiest ways to circumvent a biometric recognition system are so-called presentation attacks, in which artefacts are presented to the sensor to either impersonate another subject or avoid being recognised. In the recent past, the vulnerabilities of biometric systems to so-called morphing attacks have been unveiled. In such attacks, biometric samples of multiple subjects are merged in the signal or feature domain, in order to allow a successful verification of all contributing subjects against the morphed identity. Being a recent area of research, there is to date no standardised manner to evaluate the vulnerability of biometric systems to these attacks. Hence, it is not yet possible to establish a common benchmark between different morph detection algorithms. In this paper, we tackle this issue proposing new metrics for vulnerability reporting, which build upon our joint experience in researching this challenging attack scenario. In addition, recommendations on the assessment of morphing techniques and morphing detection metrics are given.

Unit-Selection Attack Detection Based on Unfiltered Frequency-Domain Features.

Modern text-to-speech algorithms pose a vital threat to the security of speaker identification and verification (SIV) systems, in terms of subversive usage, i.e. generating presentation attacks. In order to distinguish between presentation attacks and bona fide authentication attempts, presentation attack detection (PAD) subsystems are of utmost importance. Until now, the vast majority of introduced spoofing countermeasures rely on speech production and perception based features. In this paper, we utilize the complete frequency band without further filterbank processing in order to detect non-smooth transitions in the full and high frequency domain caused by unit-selection attacks. For the purpose of especially detecting unit selection attacks, the applicability of Fast Fourier Transformation (FFT) and Discrete Wavelet Transformation (DWT) is examined regarding non-smooth transitions in the full and high frequency domain, excluding filter-bank analyses. Gaussian Mixture Model (GMM) and Support Vector Machine (SVM) classifiers are trained on the German Speech Data Corpus (GSDC) and validated on the standard ASVspoof 2015 corpus resulting in EERs of 7.1% and 11.7%, respectively. Despite language and data shifts, the proposed unit-selection PAD scheme achieves promising biometric performance and hence, introduces a new direction to voice PAD.

Predicting the Vulnerability of Biometric Systems to Attacks based on Morphed Biometric Information

Morphing techniques can be used to create artificial biometric samples or templates, which resemble the biometric information of two or more individuals in signal and feature domain. If morphed biometric samples or templates are infiltrated to a biometric recognition system, the subjects contributing to the morphed sample can be both successfully verified against a single enrolled template. Hence, the unique link between individuals and their biometric reference data is not warranted. This leads to serious security gaps in biometric applications, in particular, the issuance and verification process of electronic travel documents. Recently, different biometric systems have been attacked using morphed biometric samples. However, so far a systematic approach to predict the vulnerability of the system to such attacks has not been proposed. In this work, the authors present a framework to evaluate the vulnerability of biometric systems to attacks using morphed biometric information. Based on a biometric systems mated/non-mated score distribution and its decision threshold, a theoretical vulnerability assessment is proposed. In an experimental evaluation, the vulnerability of a face and an iris recognition system is quantified based on the presented framework. Obtained results are verified against real attacks based on morphed face images and morphed iris-based templates.

Detecting Morphed Face Images Using Facial Landmarks.

With the widespread deployment of automatic biometric recognition systems, some security issues have been unveiled. In particular, face recognition systems have been recently shown to be vulnerable to attacks carried out with morphed face images. Such synthetic images can be defined as the fusion of the face images of two (or more) different subjects. The associated risk lies on the ability of multiple subjects to be positively verified with a single enrolled morphed face image. As common texture based features have limited capabilities to tackle this problem, we propose a novel method for morphed face image detection, based on the computation of the differences between the landmarks of a probe bona fide (i.e., captured under supervision) image of the attacker, and the landmarks of the enrolled image (i.e., the suspected morphed image). In this work, a new database is created for the experiments, comprising both bona fide and morphed images created with two different morphing methods. The experiments show that for the detection task, the proposed algorithm achieves Equal Error Rates at 32.7%.

Morph Deterction from Single Face Image: a Multi-Algorithm Fusion Approach

The vulnerability of face, fingerprint and iris recognition systems to attacks based on morphed biometric samples has been established in the recent past. However, so far a reliable detection of morphed biometric samples has remained an unsolved research challenge. In this work, we propose the first multi-algorithm fusion approach to detect morphed facial images. The FRGCv2 face database is used to create a set of 4,808 morphed and 2,210 bona fide face images which are divided into a training and test set. From a single cropped facial image features are extracted using four types of complementary feature extraction algorithms, including texture descriptors, keypoint extractors, gradient estimators and a deep learning-based method. By performing a score-level fusion of comparison scores obtained by four different types of feature extractors, a detection equal error rate (D-EER) of 2.8% is achieved. Compared to the best single algorithm approach achieving a D-EER of 5.5%, the D-EER of the proposed multi-algorithm fusion system is al- most twice as low, confirming the soundness of the presented approach.

Binarization of spectral histogram models: An application to efficient biometric identification

Feature extraction techniques such as local binary patterns (LBP) or binarized statistical image features (BSIF) are crucial components in a biometric recognition system. The vast majority of relevant approaches employs spectral histograms as feature representation, i.e. extracted biometric reference data consists of sequences of histograms. Transforming these histogram sequences to a binary representation in an accuracy-preserving manner would offer major advantages w.r.t. data storage and efficient comparison. We propose a generic binarization for spectral histogram models in conjunction with a Hamming distance-based comparator. The proposed binarization and comparison technique enables a compact storage and a fast comparison of biometric features at a negligible cost of biometric performance (accuracy). Further, we investigate a serial combination of the binary comparator and histogram model-based comparator in a biometric identification system. Experiments are carried out for two emerging biometric characteristics, i.e. palmprint and ear, confirming the soundness of the presented technique.