LBRT: Local-Information-Refined Transformer for Image Copy-Move Forgery Detection

Area of Science:

• Computer Vision
• Digital Image Forensics
• Artificial Intelligence

Background:

• Current deep learning methods for copy-move forgery detection (CMFD) often lose critical detail information.
• Existing methods struggle with extracting long-range dependencies and local tampered region details.

Purpose of the Study:

• To develop an advanced deep learning model for copy-move forgery detection.
• To address the limitations of existing methods in capturing both global context and local details.

Main Methods:

• A novel dual-branch network, the Local Branch Refinement Transformer (LBRT), was designed.
• LBRT employs Transformer encoding on both global and locally re-segmented image patches.
• Features from global and local branches are fused, followed by up-sampling and decoding.

Main Results:

• LBRT effectively models global semantic information and refines local detail information.
• Experimental results demonstrate superior performance of LBRT compared to state-of-the-art CMFD methods.
• The model achieved high accuracy on the USCISI, CASIA CMFD, and DEFACTO CMFD datasets.

Conclusions:

• The proposed LBRT network offers a significant advancement in copy-move forgery detection.
• Integrating global and local feature refinement enhances the robustness and accuracy of CMFD systems.
• LBRT provides a promising direction for future research in image forensics.