Deepfake Detection: Distinguishing Real Vs. AI-Generated Faces Via Adversarial Features

This paper presented an adversarial feature based deepfake detection framework combining spatial and frequency domain analysis. The proposed system achieves high accuracy and robustness in distinguishing real and AI-generated faces. Future work includes incorporating transformer based architectures for improved global context modeling, integrating temporal consistency analysis for video deepfake detection, exploring diffusion model specific artifact detection, and developing watermark based authentication systems for synthetic media identification.

[1]     Y. Chen et al., “Frequency Aware Deepfake Detection Using Spectral Learning,” IEEE Transactions on Information Forensics and Security, 2023. [2]    M. Kumar et al., “Transformer Based Face Forgery Detection with Global Attention,” IEEE Transactions on Artificial Intelligence, 2024. [3]   L. Zhang et al., “Adversarially Robust Deepfake Detection Networks,” IEEE Transactions on Neural Networks and Learning Systems, 2024.  [4]       R. Patel et al., “Hybrid Spatial Frequency Learning for AI-Generated Image Detection,” IEEE Access, 2025. [5]    H. Lee et al., “Explainable Deepfake Detection Using Saliency Based Analysis,” IEEE Access, 2025. [6]   A. Singh et al., “Lightweight Deepfake Detection Models for RealTime Applications,” IEEE Internet of Things Journal, 2026. [7]      T. Nguyen et al., “Cross Dataset Generalization in Deepfake Detection Systems,” IEEE Transactions on Artificial Intelligence, 2023. [8]   P. Wang et al., “Robust Face Forensics Using Deep Neural Networks,” IEEE Access, 2024. [9]      K. Roy et al., “Spectral Artifact Analysis for Synthetic Image Identification,” IEEE Access, 2025. [10]   J. Brown et al., “Media Authentication Frameworks for AI-Generated Content,” IEEE Access, 2026.