Work

Estimation and Concealment Deep Fake Detection in Images using Hybrid LSTM

Year: 2023
Type: article
Abstract: The use of deepfake techniques—wherein artificial intelligence (AI) creates influential films of actors acting out fictitious scenarios—could significantly alter how internet users evaluate the veracity of content they encounter. Given that deepfakes may be used maliciously as a source of disinformation, manipulation, harassment, and persuasion, the quality of public discourse and the protection of human rights may be impacted by content creation and modification technology. Detecting falsified media is an ever-evolving, technically complex problem that calls for teamwork from throughout the IT sector and beyond. In the Existing works, DFDC database achieve worse results and more Error Rate occurs, so to overcome this Proposed work is introduced. The proposed work aims to build innovative new technologies that can help detect deepfakes and manipulated media. The deep fake videos were generated using Generative Adversarial Network (GAN) and classified using Hybrid Long Short-Term Memory with Extreme Machine Learning Techniques (HLSTM-ELM). GAN replaces the actual image or video of the person with fake data. The suggested HLSTM-ELM brings out better classification accuracy at a lower computational cost. The comparison of the proposed technique with several Deepfake datasets that obtained results rapidly and with a performance that is better than Existing methods, including an accuracy of 93.84% on the FaceForecics++ dataset, 93.85% on the DFDC dataset, 93.66% on the VDFD dataset, and 93.43% on the Celeb-DF dataset. Our findings suggest that Proposed HLSTM-ELM techniques may be used to construct an efficient system for identifying Deepfakes.
Language: English
Open Access status: diamond
OpenAlex ID: W7217348327
FWCI: 0.384
Cites: 44
Cited by: 6
Related to: 10