Jul 2026· IEEE Transactions on Pattern Analysis and Machine Intelligence· Vol PP· 0 citations· 224 references
MedicineComputer Science
Abstract
Face de-identification (De-ID) aims to remove or conceal personally identifiable facial features in images or videos to prevent identity recognition while preserving utility for downstream tasks. With the rising emphasis on data privacy and responsible AI, face De-ID has emerged as an active research area spanning computer vision and privacy-preserving communities. Early approaches, and many contemporary ones, operate in the digital domain by modifying pixel-level or appearance-level features through post-capture processing. Recent advances extend face De-ID beyond post-processing by integrating privacy mechanisms directly into sensors during image acquisition, bridging sensing systems and downstream vision algorithms. In parallel, physical-domain methods explore wearable accessories and materials that conceal identity information in real-world environments prior to capture. In this survey, we present the first unified overview that spans the full data acquisition pipeline, encompassing the physical, sensor, and digital domains. Through this domain-centric lens, we systematically analyze current methodologies, technical progress, and the distinct challenges inherent to each stage. We then review and organize existing evaluation protocols, examining current practices and highlighting the critical need for standardized, comprehensive benchmarks. Finally, we identify key open problems and outline emerging research directions to guide future work in this rapidly evolving field. To support ongoing research, we maintain a project page that organizes relevant literature with collected datasets and open source code: https://github.com/CV-AC/Awesome-FaceDe-ID.
Soft-biometric attributes such as gender, age, and ethnicity provide valuable ancillary evidence when full identity recognition is not feasible, supporting applications in forensic investigation, identity verification, surveillance, or detection of synthetic and manipulated media. Among biometric modalities, the periocular region is a robust source of soft-biometric cues, as it often remains visible when other parts of the face are occluded, a frequent condition in forensic evidence and surveillance footage, and can be captured across a wide range of acquisition conditions. In this paper, we provide a survey of demographic attribute estimation from periocular images, covering publicly available datasets, methodological trends from handcrafted descriptors to deep learning architectures, and the state of the art in gender, age, and ethnicity prediction. We discuss use cases relevant to multimedia forensics and disinformation-detection applications, including demographic filtering in surveillance footage, age verification, and the detection of demographic inconsistencies in synthetic data. We also highlight open challenges, including dataset bias, cross-domain generalisation, fairness, ethical aspects, and the lack of forensic-oriented benchmarks.
F. Alonso-Fernandez, Kevin Hernandez-Diaz, J. Bigun· 0 citations
The human hand serves as a vital biometric modality, with the palm, fingers, and knuckles commonly employed for identity recognition across a range of applications. Despite the abundance of sensitive identity cues present in hand imagery, comprehensive privacy protection, particularly full-hand de-identification, remains underexplored. To address this gap, we introduce a self-semantic-structural-guided de-identification framework that operates directly on a single hand image, eliminating the need for training or optimization. Our method builds upon a pre-trained diffusion-based inpainting model, enhanced with IP-Adapter and ControlNet to incorporate semantic and structural cues. To mitigate identity leakage, we devise a selective semantic injection strategy and a depth-based structural de-identification strategy for each branch. Additionally, to maintain background integrity, we implement a self-adaptive hand-blending technique that seamlessly integrates the de-identified hand into the original scene. Extensive evaluations across multiple datasets and recognition systems validate the method’s effectiveness in preserving image quality, usability, and de-identification fidelity.
Licheng Yan, Yifan Lyu, Weiliang Huang et al.· IEEE Transactions on Informa...· 0 citations
Face recognition is a ubiquitously used computer vision task that has a wide range of applications ranging from everyday smartphone biometrics to high-stakes security systems. Most face recognition systems rely on traditional cameras, which often suffer from limitations such as bulky form factors, high costs, and limited privacy protection. To address these limitations, lensless cameras have emerged as an alternative. Lensless cameras use thin optical encoders, enabling smaller size, lower cost, and greater design flexibility. These cameras are typically paired with reconstruction algorithms that convert raw captures into recognizable images. However, reconstructed images often contain artifacts, and the reconstruction methods struggle to generalize well to real-world conditions. Furthermore, existing face datasets do not account for the artifacts present in lensless images. To address this issue, we introduce the Lensless Face Dataset (LFD). LFD comprises 21,080 lensless raw measurements, reconstructions, and standard images of faces captured under diverse lighting, angle, and distance. Our key contributions are: (1) Real-world lensless face data: LFD focuses on capturing a diverse face dataset with varying levels of artifacts introduced under different environments; (2) In-the-wild captures: 4,976 images are captured in outdoor settings with varying intensities of natural light and different background patterns; (3) Multiple lensless devices: LFD includes face images collected from three different types of lensless cameras, each with a unique optical encoder. We use this hardware diversity to demonstrate generalization across different lensless cameras. Through comprehensive evaluations and analysis, we show that LFD effectively captures shared features and artifacts across different lensless imaging devices, making it a valuable dataset for advancing lensless face recognition.
Junho Kim, S. S. Khan, Sara Wan et al.· 0 citations
Talking-face (TF) deepfake generation synthesizes photore- alistic facial video from a static source image and an au- dio signal, producing forgeries that current image-based detectors consistently fail to identify. Unlike face-swap ma- nipulation, TF synthesis has no underlying real video from which to inherit physiological characteristics, making re- mote photoplethysmography (rPPG) a uniquely motivated detection modality for this forgery category. We propose a detection framework that extracts per-video rPPG wave- forms via RhythmFormer and trains a suite of lightweight classifiers to distinguish real from synthesized physiologi- cal signals. Evaluated on the TF subset of Celeb-DF++ un- der a strict subject-independent protocol, where test identi- ties are completely separated from training identities, our 1D ResNet achieves an AUC of 0.806 and EER of 27.8%, placing it within 2.4 points of the best published general- purpose detector (Effort, ICML 2025) while operating ex- clusively on the physiological channel. We document a con- trolled reproduction study of DeepFakesON-Phys, the rep- resentative prior rPPG detector, demonstrating degrada- tion from AUC 0.999 on legacy face-swap data to 0.622 on the TF subset of Celeb-DF++. We further show that detec- tion difficulty is strongly method-dependent: AUC ranges from 0.985 (Real3DPortrait) to 0.690 (IP-LAP) across the seven TF generators, with the ranking remaining perfectly stable across all evaluation protocols. This spread reflects an interpretable physiological property of each generator rather than evaluation noise, and constitutes the primary theoretical contribution of the work.
Missing person identification using surveillance imagery remains a challenging problem due to adverse visual conditions, limited availability of reference images, and strict privacy constraints surrounding real-world data. This paper presents a hybrid deep-learning framework for missing person identification that combines YOLO-based person detection with deep face embedding models, specifically FaceNet and ArcFace. To address ethical and privacy limitations, a manually curated composite synthetic dataset is constructed by combining publicly available crowd and in-the-wild face datasets with additional curated images to realistically emulate CCTV conditions, including low illumination, occlusions, visually similar individuals, accessories, and clothing-matched decoys, while restricting each identity to only three to four reference images. The proposed system is evaluated across five YOLO variants (v8–v12) and a wide range of cosine similarity thresholds to analyze detection sensitivity, false positive behavior, and overall identification robustness. Extensive experiments conducted on 780 group images demonstrate that the YOLO–ArcFace pipeline achieves superior performance, reaching a peak identification accuracy of 97.50% with near-zero false positives, while the YOLO–FaceNet pipeline attains a maximum accuracy of 95.42% at an optimized operating threshold but exhibits higher susceptibility to false matches. Threshold–accuracy analysis and operating point comparison reveal that ArcFace embeddings provide stronger inter-class separation and greater stability under surveillance-specific distortions. The results highlight the importance of discriminative embedding models in safety-critical identification tasks and demonstrate that reliable missing person detection is achievable even under constrained and privacy-preserving settings, with future extensions proposed toward continuous video-based tracking using multi-object tracking algorithms.
G. S. Rakshika, Babyrani Waikhom, U. Muthaiah· International Conference Com...· 0 citations
Visible-Infrared Person Re-Identification (VI-ReID) is essential for all-weather surveillance in smart city applications. Existing methods still face two critical bottlenecks. First, significant modal discrepancies between visible and infrared images lead to serious distribution shifts, causing identity-irrelevant modality information to interfere with identity semantics. Second, decoupled shared features often lack focus on discriminative local regions, which limits the model's ability to capture fine-grained details. To address these issues, we propose a novel framework termed Modality Disentanglement and Local Discriminative Enhancement (MDLDE). We first introduce a disentanglement method based on Mutual Information Minimization to minimize statistical dependence between modality-shared and modality-specific features from a probability distribution perspective. Subsequently, a Local Discriminative Attention Module is designed to adaptively focus on highly informative body parts such as head-shoulder ratio and torso patterns. By reinforcing these localized cues, the model achieves robust fine-grained representations against complex backgrounds. Extensive experiments on the SYSU-MM01 datasets demonstrate the superiority of our method. On the SYSU-MM01 All-Search mode, we achieve 73.9% Rank-1 and 69.7% mAP, outperforming current state-of-the-art approaches.
Xiaokai Liu, Fangqing Zhou, Qian Song et al.· Advances in Engineering Tech...· 0 citations