Multi-modal texture fusion network for detecting AI-generated images.

Area of Science:

• Computer Science
• Digital Forensics
• Artificial Intelligence

Background:

• The proliferation of AI-generated content necessitates robust methods for identifying synthetic media.
• Ensuring media integrity is paramount in digital forensics and combating misinformation.

Purpose of the Study:

• To develop and evaluate a novel multi-modal fusion network for enhanced detection of AI-generated images.
• To leverage complementary texture and content information for improved accuracy in synthetic image identification.

Main Methods:

• A multi-modal fusion network integrating RGB images, Local Binary Pattern (LBP) maps, and Gray-Level Co-occurrence Matrix (GLCM) representations.
• Parallel processing of input streams via a shared-weight convolutional backbone.
• Feature-level fusion to enhance the discrimination capability of the detection model.

Main Results:

• The proposed fusion framework significantly outperforms existing single-modality detection baselines.
• The method demonstrates strong generalization capabilities across various types of AI generative models.
• Experimental validation on benchmark datasets confirms the effectiveness of the multi-modal approach.

Conclusions:

• The developed multi-modal fusion network provides an effective and reliable solution for detecting AI-synthesized imagery.
• Integrating texture and content information enhances the robustness of synthetic image detection.
• The approach offers an interpretable and efficient tool for digital forensics and media integrity applications.