This paper is published in Volume-9, Issue-6, 2023
Area
Computer Science - Artificial Intelligence
Author
Lipika Chadha, Hiya Kulasrestha, Vishesh Bhargava, Varun Jindal
Org/Univ
Vellore Institute of Technology, Vellore, Tamil Nadu, India
Pub. Date
25 November, 2023
Paper ID
V9I6-1183
Publisher
Keywords
MTCNN, CNN, InceptionV3, CLAHE, Deepfake Detection, Facial Embeddings

Citationsacebook

IEEE
Lipika Chadha, Hiya Kulasrestha, Vishesh Bhargava, Varun Jindal. Improvised approach to Deepfake detection, International Journal of Advance Research, Ideas and Innovations in Technology, www.IJARIIT.com.

APA
Lipika Chadha, Hiya Kulasrestha, Vishesh Bhargava, Varun Jindal (2023). Improvised approach to Deepfake detection. International Journal of Advance Research, Ideas and Innovations in Technology, 9(6) www.IJARIIT.com.

MLA
Lipika Chadha, Hiya Kulasrestha, Vishesh Bhargava, Varun Jindal. "Improvised approach to Deepfake detection." International Journal of Advance Research, Ideas and Innovations in Technology 9.6 (2023). www.IJARIIT.com.

Abstract

Deepfakes are realistic, human-synthesized videos that are incredibly simple to produce thanks to the advent of sophisticated algorithms fueled by advances in the field of deep learning. Deepfakes are being used to create fictitious news stories about terrorism, politics, and retaliation in order to incite societal unrest. The development of efficient techniques for identifying deepfakes is imperative, given the mounting concerns surrounding them. Our work in this field offers a brand-new deep learning-based method that effectively distinguishes between authentic videos and artificial intelligence-generated phony ones. Our technique can identify deepfakes that are both reenactments and replacements. To counter the threat posed by artificial intelligence (AI), we suggest a system that makes use of AI. In order to train an InceptionV3 model to categorize films as either real or manipulated, depending on whether they have undergone any kind of alteration, this method uses an MTCNN neural network to extract frame-level information. We assess our method on a large-scale balanced and mixed data set in order to mimic real-time scenarios and improve the model’s performance on real-time data. This dataset was painstakingly created by combining multiple available datasets.