DCU-Net: a dual-channel U-shaped network for image splicing forgery detection

Area of Science:

• Digital Image Forensics
• Computer Vision
• Deep Learning Architectures

Background:

• Image splicing forgery poses a significant challenge in digital image forensics.
• Accurate detection and localization of tampered image areas are crucial for verifying image integrity.

Purpose of the Study:

• To propose a novel image tamper localization method named DCU-Net.
• To enhance the accuracy and robustness of detecting spliced image forgeries.

Main Methods:

• A dual-channel U-Net (DCU-Net) architecture is proposed, integrating an encoder, feature fusion, and decoder.
• High-pass filters extract residual information, and a dual-channel encoder processes original and residual images.
• Dilation convolution and feature fusion techniques are employed for granular feature extraction.

Main Results:

• DCU-Net demonstrates superior performance compared to state-of-the-art algorithms on Casia2.0 and Columbia datasets.
• The method accurately locates tampered areas within images.
• Experimental results confirm the model's robustness against noise and JPEG recompression attacks.

Conclusions:

• DCU-Net offers an effective solution for image splicing forgery detection and localization.
• The proposed method achieves high accuracy and robust performance in challenging forensic scenarios.