A Comparative Analysis of Camera Optics and Latent Space Projections for Deepfake Detection

DOI

10.2991/978-94-6239-610-4_34How to use a DOI?

Keywords

Deepfake Detection; Diffusion Models; Chromatic Aberration; Generalisation; Zero-Shot Detection

Abstract

The rapid shift in deepfake generation from Generative Adversarial Networks (GANs) to advanced Diffusion Models has made many traditional detection methods ineffective. As synthetic media becomes increasingly difficult to distinguish from real content, the detectors’ inability to generalize across architectures poses a significant risk. This paper addresses this issue by comparing two detection approaches: physics-based camera optics, which focuses on lens artifacts such as chromatic aberration and sensor noise that generators cannot replicate, and latent space reverse engineering, which identifies statistical anomalies by mapping images back to the generative model’s high-dimensional space. We assess both methods using the Deepfake Eval 2024 benchmark, testing them against new generator architectures and heavy social media compression. Our findings reveal a key trade-off: Latent space methods achieve high precision on familiar generators but struggle to perform well on new ones. In contrast, Camera Optics analysis remains robust across all scenarios and outperforms data-driven methods on real-world footage. We conclude that while latent space analysis is most accurate for current threats, physics-based optical constraints are essential for detecting future, unknown generative models.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - A. R. Taniya
AU  - P. Bhavani
AU  - Vinod Kaaparthi
PY  - 2026
DA  - 2026/05/05
TI  - A Comparative Analysis of Camera Optics and Latent Space Projections for Deepfake Detection
BT  - Proceedings of the First International Conference on Advances in Forensics and Cyber Technologies (ICFACT 2025)
PB  - Atlantis Press
SP  - 386
EP  - 399
SN  - 2352-538X
UR  - https://doi.org/10.2991/978-94-6239-610-4_34
DO  - 10.2991/978-94-6239-610-4_34
ID  - Taniya2026
ER  -