International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 11 Issue: 05 | May 2024
www.irjet.net
p-ISSN: 2395-0072
DEEPFAKE DETECTION USING CNN AND SELF ATTENTION Dr. Kumaraswamy S1, Muskan Bansal2, Akshay Biradar2, C Adharsh2 1Department of Computer Science and Engineering, University of Visvesvaraya College of Engineering, Bangalore,
India
2Assistant Professor, University of Visvesvaraya College of Engineering, Bangalore, India
------------------------------------------------------------------------***-------------------------------------------------------------------------
ABSTRACT-The proliferation of deepfake technology has raised concerns about the authenticity and integrity of digital content, posing significant challenges to various domains, including media, politics, and cybersecurity. In response to this emerging threat, this research paper proposes a novel deep learning model for detecting deepfake images, leveraging advanced techniques such as self-attention mechanisms and the InceptionV3 architecture. The model is trained on a large dataset comprising authentic and manipulated images and undergoes rigorous evaluation to assess its performance. The proposed model achieves promising results, demonstrating an accuracy of 85.6% on a comprehensive test dataset. Moreover, it exhibits robustness with an ROC AUC score of 0.93 and an average precision score of 0.95, indicating its effectiveness in distinguishing between genuine and manipulated images. These results underscore the potential of the proposed approach in mitigating the adverse effects of deepfake technology on society. Furthermore, this research contributes to the ongoing efforts to combat synthetic media manipulation by providing a robust and reliable tool for identifying deepfake images. The model's capability to accurately detect manipulated content can aid in maintaining the authenticity and trustworthiness of digital media, thereby safeguarding individuals and organizations against misinformation and fraudulent activities. The proposed deepfake detection model represents a significant step forward in addressing the challenges posed by synthetic media manipulation. By leveraging state-of-the-art deep learning techniques, such as self-attention mechanisms and convolutional neural networks, the model offers a reliable solution for detecting deepfake images, thereby enhancing cybersecurity and preserving the integrity of digital content in an era of increasing technological sophistication. Key Words: Deepfake detection, Machine Learning, CNN, Inception, Neural Network, Self Attention, Image Processing, Open Forensics
1.INTRODUCTION In recent years, the digital landscape has undergone a notable transformation fueled by the widespread adoption of social media platforms and the exponential growth of online content. This shift has ushered in an era of unprecedented connectivity and information dissemination, facilitated by the ubiquity of smartphones and computers. However, amidst this proliferation of digital content, a significant concern has emerged – the rise of deepfake technology. Deepfake technology harnesses the capabilities of deep neural networks, including Generative Adversarial Networks (GANs) and convolutional neural networks (CNNs), to manipulate visual and auditory content seamlessly. By training these networks on extensive datasets of authentic data, deepfake algorithms can generate hyper-realistic replicas that closely mimic the appearance, behaviour, and speech patterns of real individuals. This ability to create synthetic media that is virtually indistinguishable from genuine content has far-reaching implications for society, undermining the reliability of digital evidence, eroding trust in media sources, and exacerbating the spread of misinformation and disinformation. At the core of deepfake technology lie various manipulation techniques such as face-swapping, lip-synching, and puppetmastering. Face-swapping involves replacing a person's face with another, leading to fabricated videos that can tarnish reputations or falsely incriminate innocent individuals. Lip-synching manipulates lip movements to synchronize with altered audio tracks, further deceiving viewers. Puppet-mastering takes this deception a step further by imitating a target individual's facial expressions and gestures, often with the intent to propagate false information on social media. The escalating threat posed by deepfakes has spurred significant research efforts aimed at developing robust detection techniques. Machine learning and deep learning algorithms, particularly convolutional neural networks (CNNs), have emerged as promising tools in this endeavour, offering automated solutions for deepfake detection. CNNs, in particular, have garnered attention for their ability to extract relevant features from image and video data automatically.
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 650