Image fusion using a multi-level image decomposition and fusion method.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Image fusion combines information from multiple images to create a superior composite image.
• Visible and infrared images offer complementary information (texture vs. contours) but have different characteristics.
• Existing fusion methods often use uniform feature extraction, neglecting sensor-specific image properties.

Purpose of the Study:

• To propose a novel image fusion method for multi-type images, specifically visible and infrared.
• To address limitations of existing methods by leveraging multi-level decomposition and deep learning.
• To improve the quality and performance of fused images.

Main Methods:

• Utilized a multi-level extended approximate low-rank projection matrix learning for salient feature extraction.
• Employed a multi-level guide filter decomposition to capture texture details from visible images.
• Implemented a novel fusion strategy using a pretrained ResNet50 network for multi-level feature integration.

Main Results:

• The proposed method successfully extracts salient and texture features from both visible and infrared images.
• Deep learning fusion strategy effectively integrates multi-level features, enhancing final image quality.
• Experimental results show superior fusion performance compared to existing methods.

Conclusions:

• The novel multi-level decomposition and deep learning fusion strategy offers significant improvements in image fusion.
• The method effectively handles complementary information from different sensor types.
• Demonstrates enhanced fusion performance and potential for advanced image processing applications.