“AWAM” – A Dual-Pathway Deepfake Discriminator for JPEGs
Open Access DepositedDeepfakes are synthesized media generated artificial intelligence/machine learning (AI/ML) techniques that manipulate or fabricate visual and audio content with a high degree of realism, of events that did not occur (DHS, 2022). The advent of deepfake technology poses a formidable challenge in the cybersecurity landscape, necessitating robust detection mechanisms(Paris et al., 2021). The challenge is further compounded when deepfakes are encoded as JPEGs, a common image format that introduces compression artifacts, complicating the detection process.This praxis introduces the Advanced Warfare Against Manipulation (AWAM) system is introduced as an innovative deepfake detection tool, a discriminator, that employs a dual-pathway convolutional neural network (CNN) specifically engineered to discern authentic digital content from deepfakes, including those obscured by JPEG compression artifacts. “AWAM”, is also a local Ghanaian slang used to describe anything considered “fake” (Danie Neilson, 2022). Utilizing the comprehensive Deep Fake Detection Challenge (DFDC) dataset (Dolhansky et al., 2020), AWAM was rigorously trained and validated, achieving an accuracy of 96% on the subset dataset of approximately 5000 videos with 4500 images extracted and maintaining commendable performance of 92% accuracy on the full dataset derived from approximately 120,000 videos, with 94,147 images extracted and processed. This accuracy rate is a crucial indicator of the model's performance, especially when dealing with large and diverse datasets like the full DFDC dataset. This metric reflects the model's ability to correctly identify both 'Fake' and 'Real' instances across a comprehensive dataset. Achieving a 92% accuracy rate in such a context suggests that the Proposed2 Model, with its dual-pathway architecture and attention mechanism, is effectively distinguishing between deepfake and genuine content, even at scale. The absence of overfitting further underscores the system's robustness. This paper proves that AWAM’s innovation contributes to the body of knowledge by offering a promising solution to a critical cybersecurity issue in AI( Wang, Q., Zhang, et al., 2020). AWAM paves the way for future integration into cybersecurity frameworks, marking a significant stride in the fight against digital misinformation.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.