ECFuse: Edge-Consistent and Correlation-Driven Fusion Framework for Infrared and Visible Image Fusion

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Infrared and visible image fusion (IVIF) aims to integrate complementary information from different modalities.
• Existing methods often struggle with fusing cross-modality information and preventing texture loss.

Purpose of the Study:

• To propose a novel edge-consistent and correlation-driven fusion framework (ECFuse) for improved IVIF.
• To address challenges in texture preservation and cross-modality information fusion.

Main Methods:

• The framework employs multi-scale transformation (MST) to decompose images into base and detail layers.
• An edge-consistency fusion module maintains edge and texture coherence in detail layers.
• A correlation-driven deep learning network fuses global and local features in base layers.

Main Results:

• ECFuse demonstrated superior performance compared to conventional and deep learning methods on TNO, LLVIP, and M3FD datasets.
• The framework effectively preserves rich edges and textures in fused images.
• ECFuse significantly improved downstream infrared-visible object detection performance.

Conclusions:

• The proposed ECFuse framework offers an effective solution for infrared and visible image fusion.
• ECFuse successfully overcomes limitations of existing methods in texture preservation and cross-modality fusion.
• The framework has practical applications in enhancing object detection tasks.