With the advancement of facial recognition technology, concerns over facial privacy breaches owing to data leaks and external attacks have been escalating. Existing de-identification methods face challenges with compatibility with facial recognition models and difficulties in verifying de-identified images. To address these issues, this study introduces a novel framework that combines face verification-enabled de-identification techniques with face-swapping methods, tailored for video surveillance environments. This framework employs StyleGAN, Pixel2Style2Pixel (PSP), HopSkipJumpAttack (HSJA), and FaceNet512 to achieve face verification-capable de-identification, and uses the dlib library for face swapping. Experimental results demonstrate that this method maintains high face recognition performance (98.37%) across various facial recognition models while achieving effective de-identification. Additionally, human tests have validated its sufficient de-identification capabilities, and image quality assessments have shown its excellence across various metrics. Moreover, real-time de-identification feasibility was evaluated using Nvidia Jetson AGX Xavier, achieving a processing speed of up to 9.68 fps. These results mark a significant advancement in demonstrating the practicality of high-quality de-identification techniques and facial privacy protection in the field of video surveillance.

In recent years, anomaly detection in digital environments has become a critical research area due to issues such as spam messages and fake news, which can lead to privacy breaches, social disruption, and undermined information reliability. Traditional anomaly detection models often require specific training for each task, resulting in significant time and resource consumption and limited flexibility. This study explores the use of Prompt Engineering with Transformer-based Large Language Models (LLMs) to address these challenges more efficiently. By comparing techniques such as Zero-shot, Few-shot, Chain-of-Thought (CoT), Self-Consistency (SC), and Tree-of-Thought (ToT) prompting, the study identifies CoT and SC as particularly effective, achieving up to 0.96 accuracy in spam detection without the need for task-specific training. However, ToT exhibited limitations due to biases and misinterpretation. The findings emphasize the importance of selecting appropriate prompting strategies to optimize LLM performance across various tasks, highlighting the potential of Prompt Engineering to reduce costs and improve the adaptability of anomaly detection systems. Future research is needed to explore the broader applicability and scalability of these methods. Additionally, this study includes a survey of Prompt Engineering techniques applicable to anomaly detection, examining strategies such as Self-Refine and Retrieval-Augmented Generation to further enhance detection accuracy and adaptability.

With the rapid growth of digital services and transactions, concerns about personal information leakage and security threats have intensified. In particular, facial images stored in surveillance systems, security agency databases, and biometric authentication platforms can be exploited for identity theft, fraud, phishing, illegal marketing, or deepfake-based misinformation. These risks have increased the need for technologies that securely protect stored facial data while enabling authorized restoration when necessary. We propose a reversible face de-identification method using format-preserving encryption (FPE). Our method integrates symmetric-key FPE into two deep neural network (DNN)-based face swap models (FaceShifter and SimSwap) that separate identity and attribute information. This approach enables the generation of de-identified facial images and allows only authorized users to restore the original data. Experiments on the LFW, FFHQ, VGGFace2-HQ, and CelebA-HQ datasets showed that, on average, the FaceShifter model achieved a 98.64% de-identification success rate and 96.86% restoration rate, while SimSwap recorded 99.58% and 99.39%, respectively. Image quality was evaluated using SSIM, FID, LPIPS, PSNR, and BRISQUE, confirming that restored images closely resemble the originals. In conclusion, the proposed method provides a robust privacy-preserving solution for facial data in digital environments, balancing security and utility while supporting lawful restoration.