MSCSCC-Net: multi-scale contextual spatial-channel correlation network for forgery detection and localization of...

Area of Science:

• Digital Image Forensics
• Computer Vision
• Signal Processing

Background:

• JPEG compression artifacts can obscure digital image forgeries.
• Existing forgery detection methods struggle with JPEG compression.
• Accurate detection and localization of forgeries in compressed images remain a challenge.

Purpose of the Study:

• To develop a novel network for detecting and localizing forgeries in JPEG compressed images.
• To address the challenge of JPEG artifacts obscuring forgery evidence.
• To simultaneously remove JPEG artifacts while preserving forgery traces.

Main Methods:

• Introduced a Multi-Scale Contextual Spatial-Channel Correlation Network (MSCSCC-Net).
• Employed multi-scale mechanisms to handle varying scales of forged areas.
• Designed Contextual Spatial Correlation Module (CSCM) and Contextual Channel Correlation Module (CCCM) for feature extraction.
• Utilized fused features for coarse-to-fine forgery detection and mask generation.
• Integrated JPEG artifact removal as a network task, ensuring forgery artifact retention.

Main Results:

• MSCSCC-Net demonstrated improved forgery detection and localization performance.
• The network effectively distinguished between JPEG artifacts, forgery artifacts, and authentic regions.
• Simultaneous removal of JPEG artifacts and preservation of forgery evidence was achieved.
• Experimental results showed superior performance compared to state-of-the-art methods.

Conclusions:

• MSCSCC-Net offers a robust solution for forgery detection and localization in JPEG compressed images.
• The proposed network effectively handles scale variations and differentiates between artifact types.
• The integrated approach of artifact removal and forgery detection enhances overall system performance.