Search

3D face reconstruction algorithms from images and videos are applied to many fields, from plastic surgery to the entertainment sector, thanks to their advantageous features. However, when looking at forensic applications, 3D face reconstruction must observe strict requirements that still make unclear its possible role in bringing evidence to a lawsuit. Shedding some light on this matter is the goal of the present survey, where we start by clarifying the relation between forensic applications and biometrics. To our knowledge, no previous work adopted this relation to make the point on the state of the art. Therefore, we analyzed the achievements of 3D face reconstruction algorithms from surveillance videos and mugshot images and discussed the current obstacles that separate 3D face reconstruction from an active role in forensic applications.

The International Fingerprint liveness Detection Competition (LivDet) is an open and well-acknowledged meeting point of academies and private companies that deal with the problem of distinguishing images coming from reproductions of fingerprints made of artificial materials and images relative to real fingerprints. In this edition of LivDet we invited the competitors to propose integrated algorithms with matching systems. The goal was to investigate at which extent this integration impact on the whole performance. Twelve algorithms were submitted to the competition, eight of which worked on integrated systems., Preprint version of a paper accepted at ICB 2019

Nowadays, facial authentication systems are present in many daily life devices. Their performance is influenced by the appearance of the facial trait that changes according to many factors such as lighting, pose, variations over time and obstructions. Adaptive systems follow these variations by updating themselves through images acquired during system operations. Although the literature proposes many possible approaches, their evaluation is often left to data set not explicitly conceived to simulate a real application scenario. The substantial absence of an appropriate and objective evaluation set is probably the motivation of the lack of implementation of adaptive systems in real devices. This paper presents a facial dataset acquired by videos in the YouTube platform. The collected images are particularly suitable for evaluating adaptive systems as they contain many changes during the time-sequence. A set of experiments of the most representative self adaptive approaches recently appeared in the literature is also performed and discussed. They allow to give some initial insights about pros and cons of facial adaptive authentication systems by considering a medium-long term time window of the investigated systems performance.

The usual trend for the conventional palmvein recognition techniques is first to extract discriminative hand-crafted feature representations from the raw images, and then feed a classifier with them. Unfortunately, it is not yet clear how the effectiveness of such features may be held in case of a large user population or in environments where the variability among acquisitions of the same person may increase. In order to face with this problem, it may be considered that the use of multiple classifiers may increase the recognition performance with respect to that of the best individual classifier, and also may handle the problem of an effective feature extraction step. In this paper, we explore the ensemble classifier approach based on Random Subspace Method (RSM), where the basic feature space is derived after a preliminary feature reduction step on the source image, and compare results achieved with and without the use of hand-crafted features. Experimental results allow us concluding that this approach leads to better results under different environmental conditions.

Scientific literature lacks of countermeasures specifically for fingerprint presentation attacks (PAs) realized with non-cooperative methods; even though, in realistic scenarios, it is unlikely that individuals would agree to duplicate their fingerprints. For example, replicas can be created from finger marks left on a surface without the person’s knowledge. Existing anti-spoofing mechanisms are trained to detect presentation attacks realized with cooperation of the user and are assumed to be able to identify non-cooperative spoofs as well. In this regard, latent prints are perceived to be of low quality and less likely to succeed in gaining unauthorized access. Thus, they are expected to be blocked without the need of a particular presentation attack detection system. Currently, the lowest Presentation Attack Detection (PAD) error rates on spoofs from latent prints are achieved using frameworks involving Convolutional Neural Networks (CNNs) trained on cooperative PAs; however, the computational requirement of these networks does not make them easily portable for mobile applications. Therefore, the focus of this paper is to investigate the degree of success of spoofs made from latent fingerprints to improve the understanding of their vitality features. Furthermore, we experimentally show the performance drop of existing liveness detectors when dealing with non-cooperative attacks and analyze the quality estimates pertaining to such spoofs, which are commonly believed to be of lower quality compared to the molds fabricated with user’s consensus.

The similarities among different acquisitions of the same fingerprint have never been taken into account, so far, in the feature space designed to detect fingerprint presentation attacks. Actually, the existence of such resemblances has only been shown in a recent work where the authors have been able to describe what they called the “user-specific effect”. We present in this paper a first attempt to take advantage of this in order to improve the performance of a FPAD system. In particular, we conceived a binary code of three bits aimed to “detect” such effect. Coupled with a classifier trained according to the standard protocol followed, for example, in the LivDet competition, this approach allowed us to get a better accuracy compared to that obtained with the “generic users” classifier alone.

Recently, the use of brain activity as biometric trait for automatic users recognition has been investigated. EEG (Electroencephalography) signal is more often used in the medical field for diagnostic purposes. However, early EEG studies adopted similar signal properties and processing tools to study individual distinctive characteristics. As a matter of fact, features related mostly to a single region of the scalp were used, thus losing information on possible links among brain areas. In this work we approached the investigation of the EEG signal as possible biometric by focusing on two recent methods based on functional connectivity, which, in contrast with previous approaches, tend to estimate the complex interactions between EEG signals by measuring the time-series statistical interdependence. Thanks to their potential complementary, we explored their fusion by feature-level and match score-level approaches. Experimental results have shown a performance improvement with respect to that of the individual systems.

Systems for still-to-video face recognition (FR) are typically used to detect target individuals in watch-list screening applications. These surveillance applications are challenging because the appearance of faces change according to capture conditions, and very few reference stills are available a priori for enrollment. To improve performance, an adaptive appearance model tracker (AAMT) is proposed for on-line learning of a track-face-model linked to each individual appearing in the scene. Meanwhile, these models are matched over successive frames against stored reference stills images of each target individual (enrolled to the system) for robust spatiotemporal FR. Compared to the gallery-face-models produced by self-updating FR systems, the track-face-models (produced by the AAMT-FR system) are updated from facial captures that are more reliably selected, and can incorporate greater intra-class variations from the operational environment. Track-face-models allow selecting facial captures for modeling more reliably than self-updating FR systems, and can incorporate a greater diversity of intra-class variation from the operational environment. Performance of the proposed approach is compared with several state-of-the-art FR systems on videos from the Chokepoint dataset when a single reference template per target individual is stored in the gallery. Experimental results show that the proposed system can achieve a significantly higher level of FR performance, especially when the diverse facial appearances captured through AAMT correspond to that of reference stills.

In order to render a biometric system robust against malicious tampering, it is important to understand the different types of attack and their impact as observed by the liveness and matching scores. In this study, we consider zero-effort impostor attack (referred to as the Z-attack), nonzero-effort impostor attack such as presentation attack or spoofing (S-attack), and other categories of attack involving tampering at the template level (U- and T-attacks). In order to elucidate the impact of all possible attacks, we (1) introduce the concepts of source of origin and symmetric biometric matchers, and (2) subsequently group the attacks into four categories. These views not only improve the understanding of the nature of different attacks but also turn out to ease the design of the classification problem. Following this analysis, we design a novel classification scheme that can take full advantage of the attack-specific data characteristics. Two realisations of the scheme, namely, a mixture of linear classifiers, and a Gaussian Copula-based Bayesian classifier, turn out to outperform a strong baseline classifier based on SVM, as supported by fingerprint spoofing experiments.

Recent experiments, reported in the third edition of Fingerprint Liveness Detection competition (LivDet 2013), have clearly shown that fingerprint liveness detection is a very difficult and challenging task. Although the number of approaches is large, none of them can be claimed as able to detect liveness of fingerprint traits with an acceptable error rate. In our opinion, in order to investigate at which extent this error can be reduced, novel feature sets must be proposed, and, eventually, integrated with existing ones. In this paper, a novel fingerprint liveness descriptor named “BSIF” is described, which, similarly to Local Binary Pattern and Local Phase Quantization-based representations, encodes the local fingerprint texture on a feature vector. Experimental results on LivDet 2011 data sets appear to be encouraging and make this descriptor worth of further investigations.

Random Subspace Method (RSM) has been demonstrated as an effective framework for gait recognition. Through combining a large number of weak classifiers, the generalization errors can be greatly reduced. Although RSM-based gait recognition system is robust to a large number of covariate factors, it is, in essence an unimodal biometric system and has the limitations when facing extremely large intra-class variations. One of the major challenges is the elapsed time covariate, which may affect the human walking style in an unpredictable manner. To tackle this challenge, in this paper we propose a multimodal-RSM framework, and side face is used to strengthen the weak classifiers without compromising the generalization power of the whole system. We evaluate our method on the TUM-GAID dataset, and it significantly outperforms other multimodal methods. Specifically, our method achieves very competitive results for tackling the most challenging elapsed time covariate, which potentially also includes the changes in shoe, carrying status, clothing, lighting condition, etc.

Recent works have investigated the robustness to spoofing attacks of multi-modal biometric systems in parallel fusion mode. Contrary to a common belief, it has been shown that they can be cracked by spoofing only one bio-metric trait. Robustness evaluation of multi-modal systems in serial fusion mode has not yet been investigated, instead. Thus, the aim of this paper is to comparatively evaluate the robustness of multi-modal systems, in serial and parallel fusion modes, under spoofing attacks. In particular, we empirically investigate the vulnerability of serial and parallel fusion of face and fingerprint biometrics to real spoofing attacks. Our results show that multi-modal systems in both fusion modes are vulnerable to attacks against a single bio-metric trait. On the other hand, they show that the serial fusion mode can attain a favorable trade-off between performance, verification time, and robustness against spoofing attacks.

The term adaptive biometric systems refers to biometric recognition systems in which an algorithm aimed to follow variations of the clients appearance has been implemented. Among others, the self update algorithm is used when only one biometric is available, and is able to add to the clients gallery novel data collected during system operation, on the basis of a updating threshold: if the novel data, compared with existing template(s), provide a matching score higher than the given threshold, they are added to the gallery. In order to avoid misclassification errors, thus inserting impostors into the clients gallery, this threshold is very conservative. Self-update algorithm has shown to be effective for many biometrics. However, no work tried to explain, so far, why self-update should work, in particular when a very conservative update threshold is used (zeroFAR threshold). This is the goal of the present paper, which provides a conceptual explanation of the self update mechanism coupled with a set of experiments on a publicly available data set explicitly designed for studying adaptive biometric systems.

Recent works have shown that multimodal biometric systems can be evaded by spoofing only a single biometric trait. In this paper, we propose a method to evaluate the robustness of such systems against spoofing attacks, when score-level fusion rules are used. The aim is to rank several score-level fusion rules, to allow the designer to choose the most robust one according to the model predictions. Our method does not require to fabricate fake biometric traits, and allows one to simulate different possible spoofing attacks using the information of genuine and impostor distributions. Reported results, using data set containing realistic spoofing attacks, show that our method can rank correctly score-level fusion rules under spoofing attacks.

“Liveness detection”, a technique used to determine the vitality of a submitted biometric, has been implemented in fingerprint scanners in recent years. The goal for the LivDet 2011 competition is to compare software-based fingerprint liveness detection methodologies (Part 1), as well as fingerprint systems which incorporate liveness detection capabilities (Part 2), using a standardized testing protocol and large quantities of spoof and live fingerprint images. This competition was open to all academic and industrial institutions which have a solution for either software-based or system-based fingerprint vitality detection problem. Five submissions across the two parts of the competition resulted in successful completion. These submissions were: Chinese Academy of Sciences Institute of Automation (CASIA), Federico II University (Federico) and Dermalog Identification SystemsGmbH (Dermalog) for Part 1: Algorithms, and GreenBit and Dermalog for Part 2: Systems. Part 1 was evaluated using four different datasets. The best results were from Federico on the Digital Persona dataset with error for live and spoof detection of 6.2% and 11.61% respectively. The best overall results for Part 1 were Dermalog with 34.05 FerrFake and 11.825% FerrLive. Part 2 was evaluated using live subjects and spoof finger casts. The best results were from Dermalog with an error for live and spoof of 42.5% and 0.8%, respectively.

Recently adaptive biometric systems have received significant boost in the research community. These systems have the ability to automatically adapt/ update templates to the variation of input samples in the changing environment.

Serial, or sequential, fusion of multiple biometric matchers has been not thoroughly investigated so far. However, this approach exhibits some advantages with respect to the widely adopted parallel approaches. In this paper, we propose a novel theoretical framework for the assessment of performance of such systems, based on a previous work of the authors. Benefits in terms of performance are theoretically evaluated, as well as estimation errors in the model parameters computation. Model is analyzed from the viewpoint of its pros and cons, by mean of preliminary experiments performed on NIST Biometric Score Set 1.

Spoofing is an open-issue for fingerprint recognition systems. It consists in submitting an artificial fingerprint replica from a genuine user. Current sensors provide an image which is then processed as a “true” fingerprint. Recently, the so-called 3rd-level features, namely, pores, which are visible in high-definition fingerprint images, have been used for matching. In this paper, we propose to analyse pores location for characterizing the “liveness” of fingerprints. Experimental results on a large dataset of spoofed and live fingerprints show the benefits of the proposed approach.

A biometric system provides poor performances when the input data exhibit intra-class variations which are not well represented by the enrolled template set. This problem has been recently faced by template update techniques. The majority of the proposed techniques can be regarded as ldquoself-updaterdquo methods, as the system updates its own templates using the recognition results provided by the same templates. However, this approach can only exploit the input data ldquonearrdquo to the current templates resulting in ldquolocalrdquo template optimization, namely, only input samples very similar to the current templates are exploited for update. To address this issue, this paper proposes a ldquoglobalrdquo optimization of templates based on the graph mincut algorithm. The proposed approach can update templates by analysing the underlying structure of input data collected during the systempsilas operation. This is achieved by a graph drawn using a pair-wise similarity measure between enrolled and input data. Investigation of this novel template update technique has been done by its application to face verification, as a case study. The reported results show the effectiveness of the proposed technique in comparison to state of art self-update techniques.

The representativeness of a biometric template gallery to the novel data has been recently faced by proposing ldquotemplate updaterdquo algorithms that update the enrolled templates in order to capture, and represent better, the subjectpsilas intra-class variations. Majority of the proposed approaches have adopted ldquoselfrdquo update technique, in which the system updates itself using its own knowledge. Recently an approach named template co-update, using two complementary biometrics to ldquoco-updaterdquo each other, has been introduced. In this paper, we investigate if template co-update is able to capture intra-class variations better than those captured by state of art self update algorithms. Accordingly, experiments are conducted under two conditions, i.e., a controlled and an uncontrolled environment. Reported results show that co-update can outperform ldquoselfrdquo update technique, when initial enrolled templates are poor representative of the novel data (uncontrolled environment), whilst almost similar performances are obtained when initial enrolled templates well represent the input data (controlled environment).

Despite its importance, a few works have been proposed for fingerprint vitality detection. In this paper, we propose a novel feature for detecting the "liveness" of fingerprint images. This feature is derived from the image power spectrum, and point out the difference between "live " and "fake" images in terms of high frequency information loss. Preliminary results on a large data set show the effectiveness of the proposed measure.