Jul 2026· Information Technology and Control· 0 citations· 34 references
TL;DR
The experimental results show that this method balances visual consistency and structural rationality while enhancing identity concealment, providing a feasible path and technical support for the secure generation and compliant application of private images.
Abstract
With the continuous development of intelligent vision technology, facial image anonymization plays an increasingly important role in public privacy protection and data compliance sharing. However, traditional methods have a contradiction between identity feature removal and image naturalness preservation, and are prone to distortion when dealing with semantic consistency issues such as posture. To this end, a structure identity separation guided facial anonymous generation method is constructed on the framework of generative adversarial networks. A structural perturbation module guided by differential privacy mechanisms is designed, and an attention guided feature weakening mechanism is introduced. Additionally, a style guided and structure preserving conditional generation network is integrated. In performance testing, the proposed method achieves identity consistency scores of 0.814 and 0.759 for frontal and lateral postures, respectively, with corresponding natural scores of 4.08 and 3.95. The inference delay, number of generated graphs per unit energy consumption, and peak memory of this method are 114.8 ms, 19.2 IpJ, and 1286 MB, respectively. The experimental results show that this method balances visual consistency and structural rationality while enhancing identity concealment, providing a feasible path and technical support for the secure generation and compliant application of private images.
A dual-domain adversarial training framework is presented for high-fidelity 4D facial makeup detection and generation under complex and variable conditions. This approach employs paired generator-discriminator architectures, each responsible for either makeup or non-makeup facial sequences, integrated via explicit cross-domain feature alignment modules. Using composite loss strategies such as knowledge and adversarial loss, the system enforces consistent and realistic generation on both fronts. Temporal regularization and temporal regularization are among these methods. The sequence learning layer and the cascaded 3D convolutional layers enhance spatiotemporal modeling to achieve robust feature abstraction in facial dynamics. At the same time, by improving dynamic data, the training distribution and system adaptability have been enhanced. The extensive validation of the large-scale 4D annotated facial dataset shows significant improvements in recognition accuracy, generation stability, and structural preservation compared to traditional and deep learning standards. Quantitative analysis shows consistent resilience to severe illumination changes, occlusion, and pose variation. Ablation experiments indicate that each module is necessary, and efficiency evaluations show that the method is scalable. The results indicate that developing domain-aware alignment and hybrid loss integration techniques is beneficial for effective facial analysis in both controlled and challenging environments. This study promotes the practical application of intelligent human-computer interaction, providing various technical solutions for realistic 4D facial recognition and modeling.
Huihui Yin, Yurui Guan· International Conference on...· 0 citations
The widespread deployment of face recognition and visual analytics has made balancing privacy protection and image usability a critical challenge. Existing anonymization methods often rely on blurring or occlusion, which suppress identity but distort key non-identity attributes. To address this issue, this study proposes a reversible face anonymization framework based on multi-conditional collaborative control (RFAMCC), integrating multi-level feature-driven anonymization, identity-space deflection, context fusion, and a steganography–recovery mechanism. Experimental evaluations demonstrated that RFAMCC preserved semantic consistency under original, compressed, and cropped conditions. It achieved correlation scores of 0.312, 0.308, and 0.301, as well as face retrieval accuracies of 83.7%, 82.4%, and 80.9%, respectively. In privacy-sensitive evaluations, gender and age inference attacks attained low success rates of 21.4% and 23.1%, indicating strong resistance to attribute leakage. Overall, RFAMCC effectively balances anonymization strength and reversibility, providing a practical solution for privacy-preserving and legally compliant facial data sharing.
Yuzhen Zhang, Houfei Song· Information Technology and C...· 0 citations
Facial aging simulation has become an essential instrument in various fields, and some of them are forensic medicine, cosmetic surgery, entertainment, and mental health. Within the scope of criminal investigation, age progression approaches help in forecasting how missing persons might look like in the future. This paper presents the proposed Hybrid Age GAN framework, termed MP-CGAN (Multi-Phase Conditional Generative Adversarial Network), developed for generating realistic, progressive facial aging. The proposed model is built upon a conditional GAN backbone, incorporating identity-preserving representation, age- and gender-conditioned feature modulation via Adaptive Instance Normalization, residual aging blocks for gradual structural transformation, and self-attention mechanisms targeting aging-sensitive facial regions. A multi-scale discriminator operating at three resolutions (128×128, 64×64, 32×32) enforces realism at both global and fine-grained levels. In contrast to approaches that focus solely on architectural design, this research primarily addresses training stability as a fundamental factor in achieving realistic aging. A multi-phase progressive training strategy, supported by N_CRITIC update scheduling and label smoothing, was adopted to systematically resolve the discriminator collapse problem D_DEAD. The model was trained on the UTK Face dataset and used 5,500 balanced face images covering 11 age groups, including both genders, over 100 training epochs. Experimental results demonstrate strong performance with a Frechet Inception Distance (FID) of 27.78, Structural Similarity Index (SSIM) of 0.6580, Peak Signal-to-Noise Ratio (PSNR) of 19.72 dB, Cosine Identity Similarity (CSIM) of 0.7818, and Mean Absolute Error (MAE) of 7.09 years at the age bucket classification level.
Amina Taha Alazwe, Y. Mohammad· IEEE Jordan Conference on Ap...· 0 citations
This work proposes SRAP, which combines per-channel truncated SVD refinement with an identity-importance mask at every optimization step, and demonstrates that SRAP substantially improves protected-image fidelity across all reported metrics while maintaining competitive identity-disruption performance.
Generative diffusion models have revolutionized facial image synthesis, yet robust identity preservation in high resolution outputs remains a critical challenge. This issue is especially vital for security systems, biometric authentication, and privacy sensitive applications, where any drift in identity integrity can undermine trust and functionality. We introduce Diff-ID, a diffusion based framework that enforces identity consistency while delivering photorealistic quality. Central to our approach is a custom 210K image dataset synthesized from CelebA-HQ, FFHQ, and LAION-Face and captioned via a fine tuned BLIP model to bolster identity awareness during training. Diff-ID integrates ArcFace and CLIP embeddings through a dual cross attention adapter within a fine tuned Stable Diffusion UNet. To further reinforce identity fidelity, we propose a pseudo discriminator loss based on ArcFace cosine similarity with exponential timestep weighting. Experiments on held out and unseen faces show that Diff-ID does not exceed InstantID in raw ArcFace Face Similarity, but achieves substantially lower FID and the strongest FIQ based identity--realism trade off among the evaluated methods. We also present a unified DDIM based morphing pipeline that enables qualitative facial interpolation without per identity fine tuning. We further argue that identity preservation and photorealism should be evaluated jointly rather than in isolation, as high identity similarity alone does not guarantee realistic outputs. To make this trade off explicit, we report Face Image Quality (FIQ) as a complementary ratio based score that combines identity similarity and perceptual realism while keeping FS and FID as the primary metrics.
T. Rizwan, Sara Atito, Muhammad Awais et al.· 0 citations
The proposed DiffAttack framework significantly outperforms existing adversarial techniques, achieving a high average attack success rate of 84.86% across multiple face recognition models (e.g., FaceNet).
