IET Biometrics最新文献

Face recognition systems have been widely deployed in various applications, such as online banking and mobile payment. However, these systems are vulnerable to face presentation attacks, which are created by people who obtain biometric data covertly from a person or through hacked systems. In order to detect these attacks, convolutional neural networks (CNN)-based systems have gained significant popularity recently. CNN-based systems perform very well on intra-data set experiments, yet they fail to generalise to the data sets that they have not been trained on. This indicates that they tend to memorise data set-specific spoof traces. To mitigate this problem, the authors propose a Deep Patch-wise Supervision Presentation Attack Detection (DPS-PAD) model approach that combines pixel-wise binary supervision with patch-based CNN. The authors’ experiments show that the proposed patch-based method forces the model not to memorise the background information or data set-specific traces. The authors extensively tested the proposed method on widely used PAD data sets—Replay-Mobile and OULU-NPU—and on a real-world data set that has been collected for real-world PAD use cases. The proposed approach is found to be superior on challenging experimental setups. Namely, it achieves higher performance on OULU-NPU protocol 3, 4 and on inter-data set real-world experiments.

Most voice biometric systems are dependent on the language of the user. However, if the idea is to create an all-inclusive and reliable system that uses speech as its input, then they should be able to recognise people regardless of language or accent. Thus, this paper investigates the effects of languages on speaker identification systems and the phonetic impact on their performance. The experiments are performed using three widely spoken languages which are Portuguese, English, and Chinese. The Mel-Frequency Cepstrum Coefficients and its Deltas are extracted from those languages. Also, this paper expands the research study of fuzzy models in the speaker recognition field, using a Fuzzy C-Means and Fuzzy k-Nearest Neighbours and comparing them with k-Nearest Neighbours and Support Vector Machines. Results with more languages decreases the accuracy from 92% to 85.59%, but further investigation suggests it is caused by the number of classes. A phonetic investigation finds no relation between the phonemes and the results. Finally, fuzzy methods offer more flexibility and in some cases, even better results compared to their crisp version. Therefore, the biometric system presented here is not affected by multilingual environments.

Recent researches have shown that the texture descriptor local tri-directional patterns (LTriDP) performs well in many recognition tasks. However, LTriDP cannot effectively describe the structure of palm lines, which results in poor palmprint recognition. To overcome this issue, this work proposes a modified version of LTriDP, called line feature local tri-directional patterns (LFLTriDP), which takes into account the texture features of the palmprint. First, since palmprints contain rich lines, the line features of palmprint images, including orientation and magnitude, are extracted. The line features are more robust to variations compared to the original grayscale values. Then, the directional features are encoded as tri-directional patterns. The tri-directional patterns reflect the direction changes in the local area. Finally, the LFLTriDP features are constructed by the tri-directional patterns, orientation and magnitude features. The LFLTriDP features effectively describe the structure of palm lines. Considering that most palm lines are curved, we use the concavity as supplementary information. The concavity of each pixel is obtained using the Banana filter and all pixels are grouped into two categories. The LFLTriDP features are refined to generate two feature vectors by the concavity to enhance the discriminability. The matching scores of the two feature vectors are weighted differently in the matching stage to reduce intra-class distance and increase inter-class distance. Experiments on PolyU, PolyU Multi-spectral, Tongji, CASIA and IITD palmprint databases show that LFLTriDP achieves promising recognition performance.

The human finger is the essential carrier of biometric features. The finger itself contains multi-modal traits, including fingerprint and finger-vein, which provides convenience and practicality for finger bi-modal fusion recognition. The scale inconsistency and feature space mismatch of finger bi-modal images are important reasons for the fusion effect. The feature extraction method based on graph structure can well solve the problem of feature space mismatch for the finger bi-modalities, and the end-to-end fusion recognition can be realised based on graph convolutional neural networks (GCNs). However, this fusion recognition strategy based on GCNs still has two urgent problems: first, lack of stable and efficient graph fusion method; second, over-smoothing problem of GCNs will lead to the degradation of recognition performance. A novel fusion method is proposed to integrate the graph features of fingerprint (FP) and finger-vein (FV). Furthermore, we analyse the inner relationship between the information transmission process and the over-smoothing problem in GCNs from an optimisation perspective, and point out that the differentiated information between neighbouring nodes decreases as the number of layers increases, which is the direct reason for the over-smoothing problem. A modified deep graph convolution neural network is proposed, aiming to alleviate the over-smoothing problem. The intuition is that the differentiated features of the nodes should be properly preserved to ensure the uniqueness of the nodes themselves. Thus, a constraint term to the objective function of the GCN is added to emphasise the differentiation features of the nodes themselves. The experimental results show that the proposed fusion method can achieve more satisfied performance in finger bi-modal biometric recognition, and the proposed constrained GCN can well alleviate the problem of over-smoothing.

With the advent of ever-fast computing, real-time processing of visual data has been gaining importance in the field of surveillance. Also, automated decision-making by visual surveillance systems has been contributing to a huge leap in the capability of such systems, and of course their relevance in social security.

This special issue aimed to discuss cloud-based architectures of surveillance frameworks as a service. Such systems, especially when deployed to work in real-time, are required to be fast, efficient, and sustainable with a varying load of visual data.

Four papers were selected for inclusion in this special issue.

Wyzykowski et al. present an approach to synthesize realistic, multiresolution and multisensor fingerprints. Based in Anguli, a handcrafted fingerprint generator, they were able to obtain dynamic ridge maps with sweat pores and scratches. Then, a CycleGAN network was trained to transform these maps into realistic fingerprints. Unlike other CNN-based works, this framework is able to generate images with different resolutions and styles for the same identity. Finally, authors conducted a human perception analysis where 60 volunteers could hardly differentiate between real and high-resolution synthesized fingerprints.

Pawar and Attar address the problem of detection and localization of anomalies in surveillance videos, using pipelined deep autoencoders and one-class learning. Specifically, they used a convolutional autoencoder and sequence-to-sequence long short-term memory autoencoder in a pipelined fashion for spatial and temporal learning of the videos, respectively. In this setting, the principle of one-class classification for training the model on normal data and testing it on anomalous testing data was followed.

Tawfik Mohammed et al. describe a framework, implemented in a RAD (Rapid Application Development) paradigm, for performing iris recognition tests, based in the well-known Daugman's processing chain. They start by segmenting the iris ring using the Integro-differential operator, along with an edge-based Hough transform to isolate eyelids and eyelashes. After the normalization of the data (pseudo-polar domain), the features are encoded using 1D log Gabor kernel. Finally, the matching step is carried out using the Hamming distance.

Barra et al. describe an approach for automated head pose estimation that stems from a previous Partitioned Iterated Function Systems (PIFS)-based approach providing state-of-the-art accuracy with high computing cost and improve it by means of two regression models, namely Gradient Boosting Regressor and Extreme Gradient Boosting Regressor, achieving much faster response and an even lower mean absolute error on the yaw and roll axis, as shown by experiments conducted on the BIWI and AFLW2000 datasets.

Identity documents (or IDs) play an important role in verifying the identity of a person with wide applications in banks, travel, video-identification services and border controls. Replay or photocopied ID cards can be misused to pass ID control in unsupervised scenarios if the liveness of a person is not checked. To detect such presentation attacks on ID card verification process when presented virtually is a critical step for the biometric systems to assure authenticity. In this paper, a pixel-wise supervision on DenseNet is proposed to detect presentation attacks of the printed and digitally replayed attacks. The authors motivate the approach to use pixel-wise supervision to leverage minute cues on various artefacts such as moiré patterns and artefacts left by the printers. The baseline benchmark is presented using different handcrafted and deep learning models on a newly constructed in-house database obtained from an operational system consisting of 886 users with 433 bona fide, 67 print and 366 display attacks. It is demonstrated that the proposed approach achieves better performance compared to handcrafted features and Deep Models with an Equal Error Rate of 2.22% and Bona fide Presentation Classification Error Rate (BPCER) of 1.83% and 1.67% at Attack Presentation Classification Error Rate of 5% and 10%.

Most deep learning-based image classification models are vulnerable to adversarial attacks that introduce imperceptible changes to the input images for the purpose of model misclassification. It has been demonstrated that these attacks, targeting a specific model, are transferable among models performing the same task. However, models performing different tasks but sharing the same input space and model architecture were never considered in the transferability scenarios presented in the literature. In this paper, this phenomenon was analysed in the context of VGG16-based and ResNet50-based biometric classifiers. The authors investigate the impact of two white-box attacks on a gender classifier and contrast a defence method as a countermeasure. Then, using adversarial images generated by the attacks, a pre-trained face recognition classifier is attacked in a black-box fashion. Two verification comparison settings are employed, in which images perturbed with the same and different magnitude of the perturbation are compared. The authors’ results indicate transferability in the fixed perturbation setting for a Fast Gradient Sign Method attack and non-transferability in a pixel-guided denoiser attack setting. The interpretation of this non-transferability can support the use of fast and train-free adversarial attacks targeting soft biometric classifiers as means to achieve soft biometric privacy protection while maintaining facial identity as utility.

The use of keystroke timings as a behavioural biometric in fixed-text authentication mechanisms has been extensively studied. Previous research has investigated in isolation the effect of password length, character substitution, and participant repetition. These studies have used publicly available datasets, containing a small number of passwords with timings acquired from different experiments. Multiple experiments have also used the participant's first and last name as the password; however, this is not realistic of a password system. Not only is the user's name considered a weak password, but their familiarity with typing the phrase minimises variation in acquired samples as they become more familiar with the new password. Furthermore, no study has considered the combined impact of length, substitution, and repetition using the same participant pool. This is explored in this work, where the authors collected timings for 65 participants, when typing 40 passwords with varying characteristics, 4 times per week for 8 weeks. A total of 81,920 timing samples were processed using an instance-based distance and threshold matching approach. Results of this study provide empirical insight into how a password policy should be created to maximise the accuracy of the biometric system when considering substitution type and longitudinal effects.

Bloom filters (BFs) and homomorphic encryption (HE) are prominent techniques used to design biometric template protection (BTP) schemes that aim to protect sensitive biometric information during storage and biometric comparison. However, the pros and cons of BF- and HE-based BTPs are not well studied in literature. We investigate the strengths and weaknesses of these two approaches since both seem promising from a theoretical viewpoint. Our key insight is to extend our theoretical investigation to cover the practical case of iris recognition on the ground that iris (1) benefits from the alignment-free property of BFs and (2) induces huge computational burdens when implemented in the HE-encrypted domain. BF-based BTPs can be implemented to be either fast with high recognition accuracy while missing the important privacy property of ‘unlinkability’, or to be fast with unlinkability-property while missing the high accuracy. HE-based BTPs, on the other hand, are highly secure, achieve good accuracy, and meet the unlinkability-property, but they are much slower than BF-based approaches. As a synthesis, we propose a hybrid BTP scheme that combines the good properties of BFs and HE, ensuring unlinkability and high recognition accuracy, while being about seven times faster than the traditional HE-based approach.