FECFusion: Infrared and visible image fusion network based on fast edge convolution

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Infrared and visible image fusion aims to combine complementary information for enhanced scene detail.
• Existing deep learning methods face challenges with performance-resource imbalance and ineffective heteromodal feature fusion.

Purpose of the Study:

• To develop a novel infrared and visible image fusion algorithm (FECFusion) that balances fusion performance and computational cost.
• To improve the extraction and fusion of texture and heteromodal features.

Main Methods:

• Utilized structural re-parameterization edge convolution (RECB) with embedded edge operators for enhanced texture feature extraction.
• Employed an attention fusion module (AFM) to fuse unique and public heteromodal features.
• Optimized the network using structural reparameterization for a VGG-like architecture, improving inference speed.

Main Results:

• FECFusion demonstrated superior performance across multiple evaluation metrics compared to seven advanced algorithms on MSRS, TNO, and M3FD datasets.
• The algorithm achieved better visual results and richer scene detail information.
• FECFusion consumed fewer computational resources than existing methods.

Conclusions:

• The proposed FECFusion algorithm effectively addresses the limitations of current deep learning fusion methods.
• It offers an efficient and high-performance solution for infrared and visible image fusion.
• The VGG-like architecture enhances fusion speed without compromising performance.