DEEPSHIELD: Vision Transformer based Deepfake Image Detection Model

View at publisher Summarise with VIRAT
Authors: K. Sivaji Raj, Cathelene George, K. Abinaya
Cited by: 0
References: 8

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Deepfake generation techniques have advanced rapidly, producing manipulated facial images that replicate subtle texture patterns, illumination cues, and structural characteristics with striking realism. Traditional CNN-based detectors, which depend heavily on shallow artifact cues, struggle to capture broader contextual relationships and often fail when exposed to unfamiliar manipulation pipelines. This creates an urgent need for a detection framework capable of analyzing long-range spatial interactions, adapting to new forms of fabricated content, and providing transparent explanations suitable for real-world verification. To address this challenge, This work presents DeepShield, a hybrid Vision Transformer framework designed to model long-range spatial relationships, adapt to unseen manipulation styles, and provide transparent explanations for real-world verification. The system uses CLIP to generate semantically rich embeddings and a fine-tuned Swin Transformer to capture both localized facial anomalies and global inconsistencies. DeepShield maintains strong generalization and reduces overfitting. Integrated Grad-CAM++ visualizations enhance interpretability, while a lightweight Streamlit interface enables real-time deployment. By combining semantic context with detailed anomaly detection, DeepShield delivers high accuracy, low latency, and dependable performance for forensic analysis, media verification, and secure communication.

Fields of Study & Subjects

Digital Media Forensic Detection Generative Adversarial Networks and Image Synthesis Misinformation and Its Impacts

Publication details

Journal / source
International Conference on Image Processing and Capsule Networks (ICIPCN)
Pages
49–55
Publication date
27 January 2026
Language
EN