Spectral edge: gradient-preserving spectral mapping for image fusion

Area of Science:

• Computer Vision
• Image Processing
• Scientific Visualization

Background:

• Image fusion aims to combine information from multiple images.
• Existing methods may struggle with preserving gradient information and color fidelity.
• High-dimensional data visualization requires effective fusion techniques.

Purpose of the Study:

• To develop a novel image fusion approach for color display.
• To generate output images with gradients matching input images.
• To enable flexible mapping of N-dimensional image data to M-dimensional outputs.

Main Methods:

• Constrained contrast mapping in the gradient domain.
• Mapping the structure tensor of high-dimensional gradients to low-dimensional fields.
• Reintegration of gradient fields to form the output image.
• Utilizing RGB rendering for color constraints.

Main Results:

• A closed-form solution for constrained optimization, enabling efficient algorithms.
• Demonstrated capability to map N-D image data to M-D outputs.
• Successful application in hyperspectral remote sensing, multi-source image fusion, and medical imaging visualization.

Conclusions:

• The proposed method offers a fast, efficient, and versatile solution for image fusion.
• It accurately preserves gradient information while incorporating color constraints.
• The approach is broadly applicable across various N-D to M-D mapping scenarios.