Interactive Feature Embedding for Infrared and Visible Image Fusion.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Deep learning for infrared and visible image fusion often uses unsupervised methods with loss functions.
• Unsupervised approaches may not fully extract all vital information from source images, leading to degradation.
• Existing methods struggle with guaranteed vital information retention in image fusion.

Purpose of the Study:

• To propose a novel interactive feature embedding within a self-supervised learning framework for infrared and visible image fusion.
• To address the limitation of vital information degradation in current unsupervised fusion methods.
• To improve the extraction and retention of crucial information from source images.

Main Methods:

• Developed a self-supervised learning framework to extract hierarchical representations from source images.
• Introduced interactive feature embedding to bridge self-supervised learning and image fusion.
• Utilized a novel approach for vital information retention in the fusion process.

Main Results:

• The proposed method effectively extracts hierarchical representations.
• Interactive feature embedding successfully bridges self-supervised learning and fusion tasks.
• Qualitative and quantitative evaluations demonstrate superior performance compared to state-of-the-art methods.

Conclusions:

• The novel self-supervised framework with interactive feature embedding overcomes vital information degradation in image fusion.
• The method achieves favorable performance in infrared and visible image fusion.
• This approach offers an effective solution for retaining vital information during image fusion.