A Bilevel Integrated Model With Data-Driven Layer Ensemble for Multi-Modality Image Fusion.

Area of Science:

• Computer Vision
• Machine Learning
• Medical Imaging

Background:

• Current image fusion methods are specialized for specific tasks and struggle with diverse scenarios.
• Conventional strategies like weighted averaging and choose-max often produce undesirable artifacts in fused images.

Purpose of the Study:

• To propose a versatile, generic image fusion method using a bilevel optimization paradigm.
• To address the limitations of existing fusion approaches in handling multi-modality image fusion tasks.

Main Methods:

• A bilevel optimization framework is employed for generic image fusion.
• Alternating optimization is performed on decoupled image components.
• Adaptive integration weight maps enable flexible fusion across different modalities.

Main Results:

• The proposed method was successfully applied to infrared and visible, CT-MRI, and MRI-SPECT image fusion.
• Quantitative and qualitative analyses demonstrate the approach's effectiveness and versatility.
• The method overcomes artifacts often seen in simpler fusion techniques.

Conclusions:

• The proposed bilevel optimization method offers a robust and adaptable solution for multi-modality image fusion.
• This generic approach shows significant performance improvements over traditional methods in various challenging applications.