AI-Based Deepfake Detection System for Fake Video and Image Identification

This paper presents a fully on-device AI-based Deepfake Detection System for Android that addresses key limitations of existing approaches, including limited content scope, lack of real-time performance, and dependency on server-based processing. The proposed system achieves a validation accuracy of 93.4% across a diverse multi-source dataset comprising GAN-generated faces, Stable Diffusion artwork, Midjourney images, synthetic objects, and real photographic content. Operating at 20 frames per second without requiring network connectivity, it enables continuous real-time monitoring through a floating overlay badge that provides three-zone classification: Real, Uncertain, and AI-Detected. To improve reliability, the system incorporates a variance-aware adaptive thresholding mechanism along with hysteresis-based flip suppression, effectively reducing false positives in edited real video content while preserving high sensitivity to genuinely AI-generated material. Furthermore, the multi-model ensemble architecture offers scalability and a clear upgrade path for incorporating specialized models tailored to different content domains. Future work will focus on developing domain-specific models for environments, cartoons, and document forgery, integrating them into the ensemble framework, optimizing performance through INT8 model quantization to reduce the current 67 MB ONNX model size and enhance inference speed on mid-range devices, and expanding platform support to iOS through CoreML export.

1. Rossler, A., Cozzolino, D., Verdoliva, L., Riess, C., Thies, J., and Niessner, M. (2019). FaceForensics++: Learning to Detect Manipulated Facial Images. IEEE International Conference on Computer Vision (ICCV), Seoul, Korea, pp. 1–11. 2. Agarwal, S., Farid, H., Gu, Y., He, M., Nagano, K., and Li, H. (2020). Protecting World Leaders Against Deep Fakes. IEEE CVPR Workshop on Media Forensics. 3. Afchar, D., Nozick, V., Yamagishi, J., and Echizen, I. (2018). MesoNet: A Compact Facial Video Forgery Detection Network. IEEE International Workshop on Information Forensics and Security (WIFS). 4. Li, L., Bao, J., Zhang, T., Yang, H., Chen, D., Wen, F., and Guo, B. (2020). Face X-ray for More General Face Forgery Detection. IEEE CVPR. 5. Ojha, U., Li, Y., and Lee, Y.J. (2023). Towards Universal Fake Image Detection Exploiting Diffusion Models. IEEE CVPR. 6. Bird, J.J., and Lotfi, A. (2023). CIFAKE: Image Classification and Explainable Identification of AI-Generated Synthetic Images. IEEE Access, vol. 12. 7. Microsoft Corporation. (2024). ONNX Runtime: Cross-Platform Inference Acceleration. https://onnxruntime.ai. 8. Tan, M., and Le, Q.V. (2019). EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks. Proceedings of the 36th International Conference on Machine Learning (ICML), Long Beach, California.