A5028035125- Face recognition and analysis
- Biometric Identification and Security
- Face and Expression Recognition
- Generative Adversarial Networks and Image Synthesis
- Domain Adaptation and Few-Shot Learning
- Forensic and Genetic Research
- Facial Nerve Paralysis Treatment and Research
- User Authentication and Security Systems
- Retinal Imaging and Analysis
- Advanced Image and Video Retrieval Techniques
- Forensic Fingerprint Detection Methods
- Face Recognition and Perception
- Anomaly Detection Techniques and Applications
- Gaze Tracking and Assistive Technology
- Gait Recognition and Analysis
- Video Surveillance and Tracking Methods
- COVID-19 diagnosis using AI
- Advanced Malware Detection Techniques
- Ethics and Social Impacts of AI
- Forensic Anthropology and Bioarchaeology Studies
- Spam and Phishing Detection
- Vehicle License Plate Recognition
- Facial Rejuvenation and Surgery Techniques
- Digital Media Forensic Detection
- Law in Society and Culture
Fraunhofer Institute for Computer Graphics Research
2018-2025
Technical University of Darmstadt
2019-2024
Fraunhofer Society
2020-2021
Learning discriminative face features plays a major role in building high-performing recognition models. The recent state-of-the-art solutions proposed to incorporate fixed penalty margin on commonly used classification loss function, softmax loss, the normalized hypersphere increase power of models, by minimizing intra-class variation and maximizing inter-class variation. Marginal losses, such as ArcFace CosFace, assume that geodesic distance between within different identities can be...
Face image quality assessment (FIQA) estimates the utility of captured in achieving reliable and accurate recognition performance. This work proposes a novel FIQA method, CR-FIQA, that face sample by learning to predict its relative classifiability. classifiability is measured based on allocation training feature representation angular space with respect class center nearest negative center. We experimentally illustrate correlation between As such property only observable for dataset, we...
This article presents FRCSyn-onGoing, an ongoing challenge for face recognition where researchers can easily benchmark their systems against the state of art in open common platform using large-scale public databases and standard experimental protocols. FRCSyn-onGoing is based on Face Recognition Challenge Era Synthetic Data (FRCSyn) organized at WACV 2024. first international aiming to explore use real synthetic data independently, also fusion, order address existing limitations technology....
Deep neural networks have rapidly become the mainstream method for face recognition (FR). However, this limits deployment of such models that contain an extremely large number parameters to embedded and low-end devices. In work, we present lightweight accurate FR solution, namely PocketNet. We utilize architecture search develop a new family face-specific architectures. additionally propose novel training paradigm based on knowledge distillation (KD), multi-step KD, where is distilled from...
Recent deep face recognition models proposed in the literature utilized large-scale public datasets such as MS-Celeb-1M and VGGFace2 for training very neural networks, achieving state-of-the-art performance on mainstream benchmarks. Recently, many of these datasets, e.g., VGGFace2, are retracted due to credible privacy ethical concerns. This motivates this work propose investigate feasibility using a privacy-friendly synthetically generated dataset train models. Towards end, we utilize...
The main question this work aims at answering is: "can morphing attack detection (MAD) solutions be successfully developed based on synthetic data?". Towards that, introduces the first synthetic-based MAD development dataset, namely Synthetic Morphing Attack Detection Development dataset (SMDD). This is utilized to train three backbones where it proved lead high performance, even completely unknown types. Additionally, an essential aspect of detailed legal analyses challenges using and...
The availability of large-scale authentic face databases has been crucial to the significant advances made in recognition research over past decade. However, legal and ethical concerns led recent retraction many these by their creators, raising questions about continuity future without one its key resources. Synthetic datasets have emerged as a promising alternative privacy-sensitive data for development. synthetic that are used train models suffer either from limitations intra-class...
Over the past years, main research innovations in face recognition focused on training deep neural networks large-scale identity-labeled datasets using variations of multi-class classification losses. However, many these are retreated by their creators due to increased privacy and ethical concerns. Very recently, privacy-friendly synthetic data has been proposed as an alternative privacy-sensitive authentic comply with regulations ensure continuity research. In this paper, we propose...
Despite the widespread adoption of face recognition technology around world, and its remarkable performance on current benchmarks, there are still several challenges that must be covered in more detail. This paper offers an overview Face Recognition Challenge Era Synthetic Data (FRCSyn) organized at WACV 2024. is first international challenge aiming to explore use synthetic data address existing limitations technology. Specifically, FRCSyn targets concerns related privacy issues, demographic...
In this paper, we present a set of extremely efficient and high throughput models for accurate face verification, Mix-FaceNets which are inspired by Mixed Depthwise Convolutional Kernels. Extensive experiment evaluations on Label Face in the Wild (LFW), Age-DB, MegaFace, IARPA Janus Benchmarks IJB-B IJB-C datasets have shown effectiveness our MixFaceNets applications requiring low computational complexity. Under same level computation complexity (≤ 500M FLOPs), outperform MobileFaceNets all...
This paper presents a summary of the Masked Face Recognition Competitions (MFR) held within 2021 International Joint Conference on Biometrics (IJCB 2021). The competition attracted total 10 participating teams with valid submissions. affiliations these are diverse and associated academia industry in nine different countries. These successfully submitted 18 solutions. is designed to motivate solutions aiming at enhancing face recognition accuracy masked faces. Moreover, considered...
Face recognition is an essential technology in our daily lives as a contactless and convenient method of accurate identity verification. Processes such secure login to electronic devices or verification at automatic border control gates are increasingly dependent on technologies. The recent COVID-19 pandemic has increased the focus hygienic methods. led wide use face masks, keep under control. effect mask-wearing collaborative environment currently sensitive yet understudied issue. Recent...
Investigating new methods of creating face morphing attacks is essential to foresee novel and help mitigate them. Creating commonly either performed on the image-level or representation-level. The representation-level has been so far based generative adversarial networks (GAN) where encoded images are interpolated in latent space produce a morphed image vector. Such process was constrained by limited reconstruction fidelity GAN architectures. Recent advances diffusion autoencoder models have...
Synthetic data is emerging as a substitute for authentic to solve ethical and legal challenges in handling face data. The current models can create real-looking images of people who do not exist. However, it known sensitive problem that recognition systems are susceptible bias, i.e. performance differences between different demographic non-demographics attributes, which lead unfair decisions. In this work, we investigate how the diversity synthetic datasets compares datasets, distribution...
Face recognition has become essential in our daily lives as a convenient and contactless method of accurate identity verification. Process such verification at automatic border control gates or the secure login to electronic devices are increasingly dependant on technologies. The recent COVID-19 pandemic have increased value hygienic However, led wide use face masks, keep under control. effect wearing mask collaborative environment is currently sensitive yet understudied issue. We address...
This paper presents a summary of the Competition on Face Morphing Attack Detection Based Privacy-aware Synthetic Training Data (SYN-MAD) held at 2022 In-ternational Joint Conference Biometrics (IJCB 2022). The competition attracted total 12 participating teams, both from academia and industry present in 11 differ-ent countries. In end, seven valid submissions were submitted by teams evaluated organizers. was to at-tract solutions that deal with detecting face morphing at-tacks while...
Deep learning-based face recognition models follow the common trend in deep neural networks by utilizing full-precision floating-point with high computational costs. Deploying such use-cases constrained requirements is often infeasible due to large memory required model. Previous compact approaches proposed design special architectures and train them from scratch using real training data, which may not be available a real-world scenario privacy concerns. We present this work QuantFace...
Launched in 2013, LivDet-Iris is an international competition series open to academia and industry with the aim assess report advances iris Presentation Attack Detection (PAD). This paper presents results from fourth of series: 2020. year's introduced several novel elements: (a) incorporated new types attacks (samples displayed on a screen, cadaver eyes prosthetic eyes), (b) initiated as on-going effort, testing protocol available now everyone via Biometrics Evaluation Testing (BEAT)*...
SARS-CoV-2 has presented direct and indirect challenges to the scientific community. One of most prominent advents from mandatory use face masks in a large number countries. Face recognition methods struggle perform identity verification with similar accuracy on masked unmasked individuals. It been shown that performance these drops considerably presence masks, especially if reference image is unmasked. We propose FocusFace, multi-task architecture uses contrastive learning be able...
Iris presentation attack detection (PAD) plays a vital role in iris recognition systems. Most existing CNN-based PAD solutions 1) perform only binary label supervision during the training of CNNs, serving global information learning but weakening capture local discriminative features, 2) prefer stacked deeper convolutions or expert-designed networks, raising risk overfitting, 3) fuse multiple systems various types increasing difficulty for deployment on mobile devices. Hence, we propose...