Two-Stage Copy-Move Forgery Detection With Self Deep Matching and Proposal SuperGlue

Area of Science:

• Computer Science
• Digital Image Forensics

Background:

• Copy-move forgery is a common image manipulation technique where parts of an image are duplicated and pasted within the same image.
• Existing methods often struggle with detecting subtle forgeries or efficiently processing large images.

Purpose of the Study:

• To propose a novel, effective, and efficient two-stage framework for copy-move forgery detection.
• To improve the accuracy and robustness of detecting duplicated regions in digital images.

Main Methods:

• A two-stage framework combining a backbone self deep matching network and a Proposal SuperGlue module.
• The first stage utilizes atrous convolution, skip matching, and spatial attention for feature enrichment.
• The second stage employs a proposal selection strategy, SuperPoint, and SuperGlue for refined matching and forgery detection.

Main Results:

• The proposed framework successfully integrates end-to-end deep matching and keypoint matching.
• Experiments on public datasets demonstrate the framework's effectiveness in copy-move forgery detection.
• The method shows improved performance in identifying both pasted and source regions.

Conclusions:

• The novel two-stage framework offers a significant advancement in copy-move forgery detection.
• This approach opens new avenues for deep learning research in digital image forensics.
• The unified approach provides optimized results for detecting image manipulations.