Face Forgery Detection by 3D Decomposition and Composition Search.

Area of Science:

• Computer Vision
• Digital Forensics
• Image Processing

Background:

• Digital face manipulation poses significant risks due to sophisticated forgery techniques.
• Existing methods struggle to detect subtle manipulations as forgery signals become less pronounced.
• Image decomposition offers a promising avenue for revealing hidden manipulation details.

Purpose of the Study:

• To develop a novel 3D decomposition-based method for detecting digital face manipulation.
• To enhance the visibility of subtle forgery artifacts by disentangling image components.
• To achieve state-of-the-art performance in digital face manipulation detection.

Main Methods:

• A 3D decomposition method treating face images as interactions between 3D geometry and lighting.
• Disentangling face images into 3D shape, lighting, common texture, and identity texture components.
• Utilizing a fine-grained morphing network for pixel-accurate 3D shape prediction and a composition search strategy for architecture construction.

Main Results:

• The decomposed components effectively highlight subtle forgery artifacts.
• The proposed architecture successfully mines forgery clues from relevant components.
• The method achieves state-of-the-art performance in detecting digital face manipulation.

Conclusions:

• 3D decomposition is a powerful technique for uncovering hidden digital face manipulations.
• The novel method demonstrates superior performance in identifying sophisticated forgeries.
• This approach offers a robust solution for the growing challenge of fake media.