Results indicate that PRIVA can support practical local identity-selective face blurring for video privacy protection, although detection robustness remains important under low-light, crowded, distant, or partially occluded face conditions.
Abstract
The increasing use of vlog videos on social media creates privacy risks because third-party faces are often unintentionally recorded and distributed without consent. Existing face blurring approaches generally apply uniform anonymization to all detected faces and do not provide an identity-selective mechanism that keeps the content creator visible while blurring other individuals. This study develops PRIVA, a desktop-based selective face blurring application that runs locally without an external AI server. The proposed pipeline integrates YOLOv8n-Face-960 for face detection, MobileFaceNet for face recognition using 512-dimensional embeddings, and Deep SORT for maintaining identity consistency across video frames. Face enrollment is performed through guided multi-pose webcam capture, while video evaluation is conducted on extracted YOLO analysis frames from five real vlog-like test videos. YOLOv8n-Face-960 achieved an overall detection precision of 95.02%, recall of 89.32%, and F1-score of 92.09%. The baseline comparison showed that YOLOv8n-Face-960 achieved a higher mean detection F1-score than MTCNN, while MobileFaceNet provided a smaller and faster recognition model than FaceNet for CPU-based local inference. For correctly detected face instances, PRIVA achieved a system precision of 99.45%, recall of 98.70%, F1-score of 99.08%, and accuracy of 98.50% in determining whether faces should be blurred or kept visible. Processing performance testing showed an average analysis speed of 4.83 FPS, average export speed of 70.05 FPS, and average processing ratio of approximately 2.40 times the original video duration. These results indicate that PRIVA can support practical local identity-selective face blurring for video privacy protection, although detection robustness remains important under low-light, crowded, distant, or partially occluded face conditions.
Bi Biometric face authentication on mobile devices faces two main challenges: computational efficiency under limited resources and vulnerability to spoofing attacks using photos or videos. This study presents an optimized face recognition pipeline by integrating the MobileFaceNet model in TensorFlow Lite (TFLite) format into a React Native application, reinforced with a liveness detection module to ensure the presence of a genuine face. The developed system consists of four main stages: (1) real-time face detection using Expo Face Detector, (2) preprocessing involving cropping and resizing to 112 × 112 pixels with pixel normalization to the range of [−1, 1], (3) MobileFaceNet inference to generate 128-dimensional face embeddings, and (4) verification using cosine similarity with a threshold of 0.75. Initial testing shows that the system operates entirely on-device without relying on external servers, with inference latency suitable for real-time attendance applications. The integration of liveness detection is proposed as an additional layer of defense against presentation attacks. The main contribution of this study is an integrated, lightweight, and secure pipeline architecture that can be implemented on mid-range mobile devices.
D. Danuri, K. Kusrini· Jurnal Penelitian Rumpun Ilm...· 0 citations
Child face identification from surveillance video remains difficult because facial crops are frequently low-resolution, blurred, partially occluded, and captured under unstable illumination. Age-related facial variation further increases the difficulty of maintaining discriminative identity embeddings for children. This paper presents RCF-Net, a degradation-aware hybrid CNN–Transformer architecture that combines surveillance-oriented image degradation, dual-branch local/global feature extraction, and learnable cross-attention fusion. MTCNN is used for face detection and alignment, ArcFace supervision is used for discriminative embedding learning, and DeepSORT can optionally be integrated to improve temporal identity consistency in video streams. To address deployment concerns raised by surveillance use, the revised framework also specifies age-progression handling, latency-aware scheduling for live video, multi-camera scaling, and adversarial/spoof-risk safeguards. Experiments are conducted using public face datasets, namely VGGFace2, CASIA-WebFace, CelebA, and IMDB-WIKI, with child-oriented filtering and synthetic surveillance degradations. Compared with representative CNN, transformer, and hybrid baselines, RCF-Net achieves the best overall accuracy of 91.4% and yields the strongest robustness under low-resolution, blur, and occlusion stress tests. The results indicate that explicit degradation modeling and local-global feature fusion are complementary for surveillance-oriented child face identification.
R. Arora, Akash Pandey, Navjeet Kaur· 2026 4th International Confe...· 0 citations
Face forgery detection is crucial for preserving the security and integrity of facial data given the rapid developments in face manipulation techniques and deep generative models. Existing methods for video face forgery detection typically assume that all frames in a forged video are manipulated, while detecting partially forged videos that contain only a subset of altered frames remains challenging. To address this issue, we propose a novel framework, UVIF, that utilizes additional annotated images to provide fine-grained supervision for detecting partial forgeries in videos. UVIF employs a unified encoder and a multi-task learning paradigm to jointly model facial videos and images for boosted video face forgery detection. A 2D backbone with temporal fusion modules is employed as the unified encoder. A pseudo labeling process is designed for video frames to bridge their representations with those of static images. A video-oriented feature alignment strategy is further introduced to reduce the distribution gap between videos and images. Extensive experiments on benchmark datasets demonstrate the effectiveness of our framework, which outperforms state-of-theart methods in detecting partially forged videos while introducing no additional computational overhead. Our code is available at https://github.com/haotianll/UVIF.
Haotian Liu, Y. Liu, Guoying Zhao et al.· 0 citations
Face recognition systems remain vulnerable to presentation attacks involving printed photographs, replayed videos, and three-dimensional masks, creating a need for accurate and computationally efficient face anti-spoofing methods. This study proposes a lightweight deep learning framework based on MobileNetV2 for detecting genuine and spoofed facial presentations. The model employs transfer learning and fine-tuning and is trained and evaluated on the Large Crowd-Collected Face Anti-Spoofing Dataset (LCC-FASD), which contains genuine and spoofed facial samples captured under varying conditions. The framework integrates image preprocessing, data augmentation, and hyperparameter optimisation to improve classification performance and generalisation. Experimental results show a precision of 95.86%, a recall of 98.45%, and an F1-score of 97.88%. Training and validation analyses, confusion-matrix evaluation, and receiver operating characteristic analysis further support the stability of the classification results. Pruning and quantisation are also applied to reduce computational complexity while preserving competitive detection performance. The resulting lightweight framework is suitable for biometric authentication systems, mobile devices, online identity-verification platforms, and edge-based security applications. The study demonstrates the potential of MobileNetV2 for efficient face anti-spoofing and provides a basis for future research on advanced presentation attacks, including deepfakes and three-dimensional masks.
Spanish-language metadata / Metadatos en españolTítulo en español:Detección ligera de ataques de suplantación facial con MobileNetV2: aprendizaje por transferencia, compresión del modelo y evaluación en LCC-FASD
Resumen:Los sistemas de reconocimiento facial continúan siendo vulnerables a ataques de presentación mediante fotografías impresas, videos reproducidos y máscaras tridimensionales, lo que genera la necesidad de métodos precisos y computacionalmente eficientes para detectar la suplantación facial. Este estudio propone un marco ligero de aprendizaje profundo basado en MobileNetV2 para distinguir entre presentaciones faciales genuinas y fraudulentas. El modelo emplea aprendizaje por transferencia y ajuste fino, y se entrena y evalúa con el conjunto de datos Large Crowd-Collected Face Anti-Spoofing Dataset (LCC-FASD), que contiene muestras faciales genuinas y fraudulentas capturadas en condiciones diversas. El marco integra preprocesamiento de imágenes, aumento de datos y optimización de hiperparámetros para mejorar el desempeño de clasificación y la capacidad de generalización. Los resultados experimentales muestran una precisión del 95.86 %, una exhaustividad del 98.45 % y una puntuación F1 del 97.88 %. Los análisis de entrenamiento y validación, la evaluación mediante matriz de confusión y el análisis de la característica operativa del receptor respaldan adicionalmente la estabilidad de los resultados de clasificación. Asimismo, se aplican técnicas de poda y cuantización para reducir la complejidad computacional, al tiempo que se mantiene un desempeño de detección competitivo. El marco ligero resultante es adecuado para sistemas de autenticación biométrica, dispositivos móviles, plataformas de verificación de identidad en línea y aplicaciones de seguridad basadas en el borde. El estudio demuestra el potencial de MobileNetV2 para la detección eficiente de suplantación facial y proporciona una base para futuras investigaciones sobre ataques de presentación avanzados, incluidos los deepfakes y las máscaras tridimensionales.
Palabras Claves:detección de suplantación facial; detección de ataques de presentación; MobileNetV2; aprendizaje por transferencia; red neuronal convolucional; autenticación biométrica; LCC-FASD.
Smart citations:
https://scite.ai/reports/10.61467/2007.1558.2026.v17i4.1529Dimensions.Open Alex.
M. Abbas, Muhammad Munib, Muhammad Sajid et al.· International Journal of Com...· 0 citations
A detection framework is proposed that extracts per-video rPPG wave- forms via RhythmFormer and trains a suite of lightweight classifiers to distinguish real from synthesized physiologi- cal signals and shows that detec- tion difficulty is strongly method-dependent.
Deep fake technology has significantly advanced the creation of synthetic images and videos, sparking widespread concerns about its potential misuse in spreading misinformation, violating privacy, and enabling identity theft. As these manipulations be-come increasingly sophisticated, the development of reliable detection methods has become a pressing necessity. This research tackles this challenge by proposing a robust deep fake detection pipeline, leveraging a custom dataset created using Roboflow. The dataset is divided into two primary classes: real and fake, with the fake class further categorized into three subtypes based on complexity: easy fake, mid fake, and hard fake. Easy fake images involve basic manipulations that are easily identifiable by the human eye, while mid fake images combine AI-generated and human-generated elements, and hard fake images are entirely AI-generated, posing significant challenges for detection. To ensure authenticity and diversity, real images were collected from personal networks and online repositories. We trained and evaluated four YOLO-based models YOLOv8, YOLOv9, YOLOv10, and YOLOv11 for the detection task. YOLOv8 emerged as the top-performing model, achieving an accuracy of 96.2% in distinguishing between real and fake images. Finally, addressing ethical considerations and developing countermeasures to mitigate the societal impact of deep fakes should remain a priority for future research.
Unknown authors· Discover Computing· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
