Enhancing DenseNet networks for improved forensic image analysis and tampered image detection

Area of Science:

• Computer Science
• Digital Forensics
• Artificial Intelligence

Background:

• Sophisticated image manipulation techniques pose challenges to traditional forensic image analysis (FIA).
• Accurate and efficient detection of tampered images is crucial for law enforcement and media.
• Existing methods struggle with subtle image tampering detection.

Purpose of the Study:

• To enhance DenseNet architectures for improved tampered image detection.
• To increase accuracy, reduce processing time, and improve the robustness of tampered image detection.
• To address limitations in detecting subtle image tampering.

Main Methods:

• Utilized Gabor-bilateral filtering (G-BF) for enhanced feature extraction.
• Employed MS-DenseNet for multiscale feature extraction (MSFE) and attention mechanisms (AMs).
• Integrated GAN-DenseNet for realistic feature generation.

Main Results:

• Achieved a significant improvement in tampered image detection accuracy from 85% to 95%.
• Reduced processing time from 5-7 seconds to under 1 second.
• Demonstrated increased model robustness for real-world forensic analysis applications.

Conclusions:

• The enhanced DenseNet model offers superior performance in tampered image detection.
• The methodology provides a more robust and efficient solution for forensic image analysis.
• Future work will focus on integrating advanced attention mechanisms and optimizing for higher accuracy.