Explainable AI (XAI) in Deepfake Detection: Transparency and Interpretation in Computer Vision Models
Explainable AI (XAI) in Deepfake Detection: Transparency and Interpretation in Computer Vision Models
DOI:
https://doi.org/10.51473/rcmos.v1i1.2023.1867Keywords:
XAI; Deepfake; Interpretability; Transparency; Digital Forensics; Computer Vision; Trust.Abstract
The increasing sophistication of deepfakes has escalated the urgency for robust Deep Learning detection systems.However, the "black-box" nature of Computer Vision models, such as Convolutional Neural Networks (CNNs) and Transformers, poses a significant barrier to their acceptance in critical domains like forensics and law. This paper explores the application of Explainable Artificial Intelligence (XAI) techniques within the context of deepfakedetection, investigating how model transparency and interpretability can be achieved. We discuss post-hoc and intrinsic methodologies, such as CAMs (Class Activation Maps), SHAP, and LIME, analyzing their capacity to generate visual and logical evidence regarding the classification process, specifically identifying the regions of the image or video (artifacts) that are decisive for the forgery decision. The primary objective is to demonstrate that the integration of XAI is indispensable for building the necessary trust in detection systems, transforming the algorithmic decision into verifiable expert evidence, which is essential for establishing the validity and admissibility of these technologies in courts and investigations.
Downloads
References
BACH, S. et al. On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation. PLoS ONE, v. 10, n. 7, p. e0130140, 2015. DOI: https://doi.org/10.1371/journal.pone.0130140
GOODFELLOW, I. et al. Generative adversarial networks. In: ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS. 2014.
KINGMA, D. P.; BA, J. Adam: a method for stochastic optimization. In: INTERNATIONAL CONFERENCE ON LEARNING REPRESENTATIONS. 2015.
RIBEIRO, M. T.; SINGH, S.; RINARD, C. G. Why should I trust you?: explaining the predictions of any classifier. In: ACM SIGKDD INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY AND DATA MINING. 2016. DOI: https://doi.org/10.1145/2939672.2939778
SUNDARARAJAN, M.; TALY, A.; YAN, V. Axiomatic attribution for deep networks. In: INTERNATIONAL CONFERENCE ON MACHINE LEARNING. 2017.
VASWANI, A. et al. Attention is all you need. In: ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS. 2017.
SELVARAJU, R. R. et al. Grad-CAM: visual explanations from deep networks via gradient-based localization. In: IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION. 2017. DOI: https://doi.org/10.1109/ICCV.2017.74
LUNDBERG, S. M.; LEE, S.-I. A unified approach to interpreting model predictions. In: ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS. 2017.
FONG, R.; VEDANTAM, S.; LIM, J. H. Interpretable deep learning for image analysis. arXiv, 2019.
ROSSLER, A. et al. FaceForensics++: learning to detect manipulated facial images. In: INTERNATIONAL CONFERENCE ON COMPUTER VISION. 2019. DOI: https://doi.org/10.1109/ICCV.2019.00009
Downloads
Published
Issue
Section
Categories
License
Copyright (c) 2023 Matheus de Oliveira Pereira Paula (Autor)
This work is licensed under a Creative Commons Attribution 4.0 International License.
This work is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). This means you are free to:
- Share — copy and redistribute the material in any medium or format.
- Adapt — remix, transform, and build upon the material for any purpose, even commercially.
The use of this material is conditioned upon proper attribution to the original author(s), providing a link to the license, and indicating if changes were made. The license does not require permission from the author or the publisher as long as these conditions are met.
The Creative Commons license logo is displayed permanently in the footer of the journal.
The copyright of the manuscript can be retained by the authors without restrictions and can be requested at any time, even after publication in the journal.
How to Cite
