Frequency-Aware Deepfake Detection: A Systematic Survey of CNN, Transformer, and Spectral Approaches

Deepfake media generated using modern generative modelsposes serious threats to digital trust, enabling realistic identity manipulation and misinformation. Deepfake detection research has evolved fromhandcrafted forensic cues to deep learning architectures such as convolutional neural networks (CNNs), Vision Transformers (ViTs), andfrequency-aware methods. This survey systematically reviews frequencyaware detection techniques, benchmark datasets, evaluation protocols,and open challenges. We conclude by highlighting promising directions— frequency-spatial fusion, self-supervision, and multimodal forensic systems — for robust and explainable deepfake detection.