Enhanced MesoNet-based deepfake detection using deep learning: A robust framework for multimedia forensics.

Area of Science:

• Computer Science
• Artificial Intelligence
• Digital Forensics

Background:

• Deepfake technology, utilizing artificial intelligence, poses a significant threat to digital content authenticity and public trust.
• The need for effective deepfake detection methods is critical to combat deceptive media.
• Existing detection models require enhancement to meet the challenges posed by sophisticated deepfakes.

Purpose of the Study:

• To introduce an enhanced MesoNet convolutional neural network for improved deepfake detection.
• To evaluate the performance of the enhanced model against established deepfake detection architectures.
• To develop a real-time system for practical deepfake detection applications.

Main Methods:

• An enhanced MesoNet model was developed by incorporating two additional convolutional layers.
• The model's performance was evaluated using key metrics including precision, recall, F1-score, accuracy, and MCC.
• A real-time detection system was implemented using a React frontend and Flask backend.

Main Results:

• The enhanced MesoNet model achieved high performance metrics: 96.60% precision, 95.33% recall, 95.96% F1-score, 95.59% accuracy, and 91.11% MCC.
• The proposed model demonstrated superior performance compared to baseline models like ResNet-50, VGG, and AlexNet.
• The real-time detection system proved the model's viability for practical deployment.

Conclusions:

• The enhanced MesoNet model offers a robust and scalable solution for deepfake detection.
• This research provides a foundation for real-world applications in digital forensics and content authenticity verification.
• The study highlights the potential of deep learning advancements in addressing the challenges of manipulated digital media.