Deep fake detection and classification using error-level analysis and deep learning

Area of Science:

• Computer Science
• Artificial Intelligence
• Image Processing

Background:

• The proliferation of deep fakes on social media poses a significant threat due to the ease of creation and potential for spreading disinformation.
• Traditional machine learning methods struggle with complex patterns and data variations inherent in deep fake detection.
• A robust system is crucial for differentiating authentic content from manipulated media in the digital age.

Purpose of the Study:

• To propose and evaluate an automated method for classifying deep fake images.
• To address the limitations of traditional machine learning in handling complex image manipulations.
• To develop a reliable system for detecting deep fakes and mitigating their harmful effects.

Main Methods:

• Employed a framework combining Error Level Analysis (ELA) for initial modification detection.
• Utilized Convolutional Neural Networks (CNNs) for deep feature extraction from images.
• Classified features using Support Vector Machines (SVM) and K-Nearest Neighbors (KNN) with hyper-parameter optimization.

Main Results:

• The proposed method achieved a highest accuracy of 89.5% using a Residual Network and K-Nearest Neighbor classifier.
• Demonstrated the efficiency and robustness of the automated deep fake detection approach.
• Validated the system's capability to generalize to unseen data and handle variations.

Conclusions:

• The developed deep learning and machine learning-based system effectively detects deep fake images.
• The proposed technique offers a robust solution for combating disinformation and propaganda spread through manipulated media.
• This automated method can be deployed to enhance content authenticity verification on social media platforms.