A Novel Counterfeit Feature Extraction Technique for Exposing Face-Swap Images Based on Deep Learning and Error Level...

Area of Science:

• Computer Vision
• Digital Forensics
• Artificial Intelligence

Background:

• Deep learning significantly enhances face-swap image generation, making DeepFake manipulations difficult to detect.
• Current detection methods struggle with the realism of advanced DeepFake technology.

Purpose of the Study:

• To develop an efficient technique for distinguishing DeepFake images from real facial images.
• To improve the accuracy and reduce the computational cost of counterfeit image detection.

Main Methods:

• A novel counterfeit feature extraction technique combining deep learning and Error Level Analysis (ELA).
• Utilizing the distinct compression ratios between foreground (fake face) and background (original area) in DeepFake images.
• Employing Convolution Neural Networks (CNNs) for counterfeit feature extraction and fake image detection.

Main Results:

• The Error Level Analysis (ELA) method effectively detects differences in image compression ratios.
• The proposed technique significantly improves the training efficiency of CNN models.
• Accurate extraction of counterfeit features leading to superior simplicity and efficiency compared to direct detection methods.

Conclusions:

• The developed technique offers a highly efficient and accurate method for DeepFake image detection.
• Significant reduction in computational requirements (over 90% reduction in floating-point operations) without compromising accuracy.
• The integration of ELA with deep learning provides a robust solution for identifying sophisticated face-swap manipulations.