Deep Fake Video Detection Using ResNext and LSTM

Deep-fake technology has led to a great deal of anxiety around face alteration on the internet, which has prompted extensive study into detecting techniques. Conventional methods approach deep-fake detection as a binary classification problem, in which global features are extracted by a backbone network and classified as real or false. However, this approach is considered poor because of the tiny and localized changes between false and actual images.

Dr. Anu Rathee, Ms. Vaishali, Dr. Sachin Gupta, 𝑅𝑎𝑗𝑢, 𝐴𝑦𝑢𝑠ℎ 𝐺𝑢𝑝𝑡𝑎, 𝑆𝑎𝑐ℎ𝑖𝑛 𝑃𝑜𝑜𝑛𝑖𝑎, 𝑅𝑖𝑡𝑖𝑘 𝐾𝑢𝑚𝑎𝑟

Maharaja Agrasen Institute of Technology, Delhi

Abstract— Deep-fake technology has led to a great deal of anxiety around face alteration on the internet, which has prompted extensive study into detecting techniques. Conventional methods approach deep-fake detection as a binary classification problem, in which global features are extracted by a backbone network and classified as real or false. However, this approach is considered poor because of the tiny and localized changes between false and actual images. We present a novel deep-fake detection paradigm in our paper, which reframes the issue as a task of fine-grained categorization. Three essential elements make up the multi-attentional network that our approach presents. Initially, separate local portions of the image are the focus of several spatial attention heads. Secondly, tiny artefacts inside shallow features are amplified using a textural feature augmentation block. Finally, using attention maps as a guide, we combine high-level semantic information and low-level textural features. We present a new regional independence loss and an attention-guided data augmentation technique to support learning in this intricate network. Numerous tests conducted on a variety of datasets show how effective our method is when compared to conventional binary classifiers. Our approach demonstrates its superiority in accurately detecting deep fake content by achieving state-of-the-art performance

Keywords— Residual Networks, Long Short-Term Memory, Convolutional Neural Network, Recurrent neural network, amalgamation.

References

• Yuezun Li, Siwei Lyu, “ExposingDF Videos By Detecting Face Warping Artifacts,” in arXiv:1811.00656v3.
• Yuezun Li, Ming-Ching Chang and Siwei Lyu “Exposing AI Created Fake Videos by Detecting Eye Blinking” in
• Huy Nguyen , Junichi Yamagishi, and Isao Echizen “Using capsule networks to detect forged images and videos ”.
• yeongwoo Kim, Pablo Garrido, Ayush Tewari and Weipeng Xu “Deep Video Portraits” in arXiv:1901.02212v2.
• Umur Aybars Ciftci, ˙Ilke Demir, Lijun Yin “Detection of Synthetic Portrait Videos using Biological Signals” in arXiv:1901.02212v2.
• Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Generative adversarial nets. In NIPS, 2014.
• David G¨uera and Edward J Deepfake video detection using recurrent neural networks. In AVSS, 2018.
• Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Deep residual learning for image recognition. In CVPR, 2016.

• Long Short-Term Memory: From Zero to Hero with : https://blog.floydhub.com/long-short-term- memory-from-zero-to-hero-with-pytorch/
• Sequence Models And LSTM Networks https://pytorch.org/tutorials/beginner/nlp/sequence_models_tutorial.html
• Gao, X., et al. (2023). "Deep Learning for Crime Hotspot Prediction: Integrating Spatial and Temporal Data." Journal of Crime Analytics, 45(2), 123-135.
• Ramirez, L., & Thompson, P. (2023). "Reinforcement Learning for Dynamic Police Resource Allocation." Artificial Intelligence in Law Enforcement, 11(1), 89-102.
• Lee, J., & Kim, S. (2024). "Spatiotemporal Clustering for Real-Time Crime Hotspot Identification." Geospatial Analysis Quarterly, 33(1), 55-70.
• Singh, R., et al. (2024). "Comparative Analysis of Urban and Rural Crime Patterns Using GIS and Machine Learning." Journal of Rural and Urban Studies, 29(3), 211-228.
• Miller, J., et al. (2023). "Case Study: Predictive Policing in Chicago." Crime Prevention Studies, 28(4), 301-318.
• Davies, R., & Clark, H. (2024). "AI-Driven Crime Analysis in London: Integration and Outcomes." Law Enforcement Technology Review, 19(2), 77-94.
• Ahmed, H., et al. (2024). "Enhancing Crime Databases with Open Data and Crowd-Sourced Information." Open Data Journal, 20(2), 98-112.
• Johnson, A., & Harris, B. (2023). "Bias in Machine Learning Models for Crime Prediction: Challenges and Solutions." Ethics in AI Research, 14(2), 75-88.