SFPFusion: An Improved Vision Transformer Combining Super Feature Attention and Wavelet-Guided Pooling for Infrared...

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Convolutional Neural Networks (CNNs) excel at local feature extraction but have limited receptive fields.
• Transformer architectures capture global features but often neglect crucial image details.
• Current image fusion methods struggle to balance global feature extraction with detail enhancement.

Purpose of the Study:

• To develop a novel image fusion network, SFPFusion, that enhances detail preservation.
• To improve infrared and visible image fusion by integrating global and multi-scale feature extraction.
• To address the limitations of existing Transformer-based fusion methods in detail enhancement.

Main Methods:

• Introduced a Super Feature Attention mechanism for long-range dependency modeling and global feature extraction.
• Employed Wavelet-Guided Pooling to extract multi-scale base information and enhance fine details.
• Utilized simple fusion strategies enabled by the network's powerful feature representation.

Main Results:

• SFPFusion effectively extracts both global and multi-scale features, significantly enhancing detail preservation.
• Qualitative and quantitative experiments demonstrate superior performance compared to state-of-the-art methods.
• The proposed method achieves better fusion results on multiple image fusion benchmarks.

Conclusions:

• SFPFusion offers a robust solution for infrared and visible image fusion by prioritizing detail enhancement.
• The integration of Super Feature Attention and Wavelet-Guided Pooling advances the capabilities of fusion networks.
• This approach provides a strong foundation for future research in detail-aware image fusion.