Privacy-preserving face data synthesis for digital forensics applications using deep generative models

Data privacy in digital forensics is increasingly critical, particularly in light of data protection frameworks such as Indonesia’s Data Protection Law No. 27/2022. The use of real biometric facial data poses significant risks related to identity exposure and legal compliance. To address this, we propose a privacy-preserving approach that leverages Deep Convolutional Generative Adversarial Networks (DCGANs) to generate synthetic facial images from anonymized institutional datasets. The DCGAN model was trained using adversarial learning, producing visually realistic images with improved convergence characteristics compared to a baseline f-divergence Generative Adversarial Networks (FGAN) model. Specifically, DCGAN achieved a lower generator loss (0.3 vs. 0.9) and higher discriminator accuracy (0.9 vs. 0.1). Forensic analysis using tools such as Principal Component Analysis (PCA), error level analysis, and metadata inspection identified synthetic traits in 66.5% of 200 generated images. A downstream gender classification task achieved 63.72% accuracy for female and 36.28% for male classifications, indicating that the synthetic data retains sufficient semantic structure for analytical use. These fin dings demonstrate that synthetic face data generated via deep generative models can support forensic analysis while aligning with data privacy and regulatory requirements. The proposed method offers a scalable, ethical, and legally compliant pathway for the integration of synthetic biometric data into digital forensic workflows.