Image Copy-Move Forgery Detection via Deep PatchMatch and Pairwise Ranking Learning.

Area of Science:

• Computer Vision
• Digital Forensics
• Machine Learning

Background:

• Deep learning models show promise in copy-move forgery detection (CMFD).
• Existing methods struggle with generalizability due to variations in training data and background blending.
• Convolutional operations in current deep models are insufficient for distinguishing subtle forged regions.

Purpose of the Study:

• To develop a novel, end-to-end CMFD framework integrating conventional and deep learning techniques.
• To improve the generalizability and accuracy of CMFD algorithms in practical scenarios.
• To address limitations of existing methods in detecting forgeries with background blending.

Main Methods:

• A deep cross-scale PatchMatch (PM) method customized for CMFD was developed to locate copy-move regions.
• Features from high-resolution scales are utilized for explicit, reliable point-to-point matching.
• A novel pairwise rank learning framework was proposed to effectively discriminate between source and target regions, even with background blending.

Main Results:

• The proposed framework demonstrates remarkable generalizability across diverse copy-move forgery scenarios.
• It significantly outperforms existing CMFD methods in experimental evaluations.
• The method successfully identifies subtle differences and discriminates forged regions effectively.

Conclusions:

• The integrated framework offers a robust solution for copy-move forgery detection.
• The approach enhances the reliability of CMFD in real-world applications.
• This study advances the state-of-the-art in digital image forensics.