International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 03 | Mar 2024
p-ISSN: 2395-0072
www.irjet.net
Unmasking Deepfakes: A Deep Learning Approach for Accurate Detection and Classification of Synthetic Videos SK. Abdul Sattar1,T. Guru Preetham2, V.Kalyan3,P.Venu4,B.Avinash5 1,2,3,4B.Tech. Students, Department of IT, Vasireddy Venkatadri Institute of Technology, Guntur 5Assistant Professor, Department of IT, Vasireddy Venkatadri Institute of Technology, Guntur ---------------------------------------------------------------------***---------------------------------------------------------------------
ABSTRACT:
train our model on a diverse dataset called Celeb-DF , containing various examples of manipulated videos. We also prioritize accessibility by creating a user-friendly interface with ReactJs and Flask, ensuring that individuals can easily utilize our deepfake detection system to protect against the dangers of deceptive media.
The growth of deepfakes fuels the spread of misinformation, undermining trust in media and information sources. Additionally, they worsen societal divisions by sharing fake content, leading to confusion and polarization. Deepfakes are becoming increasingly common, making it harder to spot them because they look so real. This paper addresses this problem by introducing a method to detect differences in facial features during video creation. Detection of deepfakes can be tricky due to their high realism, but our approach helps identify these fake videos by spotting changes in facial structures. Our model employs a Res-Next Convolutional Neural Network to extract frame-level features, which are then utilized to train a Long Short-Term Memory (LSTM)-based Recurrent Neural Network (RNN). This RNN classifies videos whether they are subjected to any manipulation or not. We have used a dataset called "CelebDF" to train our model to detect differences created around the face during deepfake construction. Integrated with a user-friendly interface utilizing ReactJs on the front end and Flask on the backend, our solution ensures robust defense against potential threats posed by deepfakes while prioritizing accessibility and usability.
2. LITERATURE REVIEW: The widespread emergence of deepfake videos and their misuse poses a serious threat to democracy, justice, and public trust. As a result, there is a growing need for improved methods to analyze, detect, and intervene in fake videos. Employing 26 distinct deep convolutional models, they proposed a technique to enhance detection precision, despite potential computational hurdles in managing and fine-tuning multiple models [1]. Introducing an innovative method, they utilize Cascaded Deep Sparse Auto Encoder (CDSAE) with temporal CNN for feature extraction, aiming to boost accuracy, although facing limitations due to the intricate nature of deepfake techniques [2].
Key Words: DeepFake Video, Res-Next Convolution Neural Network, Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), Flask
Relying on feature fusion with MesoInception, their model extracts deep characteristics from various iterations of target faces, showcasing effectiveness in identifying altered faces in videos, albeit with performance variations depending on the complexity of facial manipulation techniques [3].
1. INTRODUCTION: This paper tackles the growing threat posed by deepfake videos, which convincingly alter reality and can be used to deceive or manipulate viewers. These videos, capable of fabricating events or speeches, present serious risks in areas such as politics and personal privacy. To address this issue, we propose a straightforward method to detect deepfakes by examining changes in facial features during video creation.
Their approach suggests merging Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) to expedite training by leveraging pre-trained CNN models while enhancing detection precision through combined CNN and RNN architectures, though encountering challenges in adapting to new deepfake variations [4].
Our approach focuses on spotting subtle differences in facial expressions that distinguish genuine footage from manipulated content. Using advanced technologies like neural networks, specifically a combination of Res-Next CNN and LSTM-based RNN, our system can effectively identify deepfakes with a high level of accuracy. Additionally, we
© 2024, IRJET
|
Impact Factor value: 8.226
Presenting the YOLO-CNN-XGBoost strategy, they leverage YOLO face detection for precise face region extraction in video frames, backed by XGBoost for thorough deepfake detection. However, challenges arise due to reliance on pretrained models, possibly limiting effectiveness against emerging deepfake types [5].
|
ISO 9001:2008 Certified Journal
|
Page 724