FDAE: Lightweight Privacy Protection Based on Face Detection and Image Encryption

The increasing prevalence of facial privacy leakage in digital images has become a growing concern. To address this issue, we propose Face Detection and Encryption (FDAE), a lightweight privacy protection framework. FDAE integrates enhanced face detection with targeted encryption, offering an efficient alternative to full-image encryption. Specifically, we developed a YOLOv7-based face detection algorithm, further enhanced with Explicit Visual Center (EVC) and Spatial Context Pyramid (SCP) modules, to accurately localize facial regions. This approach achieves detection accuracies of 96.4% (Easy), 94.8% (Medium), and 90.3% (Hard) categories, representing improvements of 1.3%, 1.4%, and 4.8% over the baseline YOLOv7 model, respectively. Subsequently, a lightweight encryption method, which combines Logistic and PWLCM chaotic systems with DNA encoding, is applied to the detected facial regions. To further improve efficiency, we encrypt only the upper four bit-planes of the detected face regions after bit-plane decomposition. Experimental results demonstrate that, using a 44.42M-parameter lightweight model, the FDAE algorithm achieves significantly higher efficiency in facial region encryption compared to full-image encryption, with encryption and decryption speeds improved by 7.64 to 9 times. Moreover, the decrypted images exhibit negligible differences from the original ones, making FDAE well-suited for batch image processing tasks and practical privacy protection applications.