Multispectral Image Stitching via Global-Aware Quadrature Pyramid Regression.

Area of Science:

• Computer Vision
• Image Processing
• Remote Sensing

Background:

• Visible image stitching struggles in adverse environmental conditions.
• Infrared imaging offers superior penetration and environmental resilience.
• Multispectral approaches are needed for robust, all-weather scene perception.

Purpose of the Study:

• To develop an infrared and visible image-based multispectral image stitching method.
• To achieve all-weather, broad field-of-view (FOV) scene perception.
• To enhance panorama generation robustness and credibility.

Main Methods:

• Utilizing salient structural information from infrared and textual details from visible images.
• Employing a multiscale progressive mechanism with quadrature correlation for cross-modality feature correspondence.
• Integrating deformation parameters from both modalities for accurate homography estimation.
• Implementing a global-aware guided reconstruction module for seamless image fusion.

Main Results:

• The proposed method outperforms traditional cascaded fusion-stitching.
• Achieved more robust and credible panorama generation.
• Demonstrated superior performance in qualitative and quantitative evaluations.

Conclusions:

• Infrared and visible image fusion enables robust all-weather panorama perception.
• The developed method effectively leverages complementary multispectral information.
• The approach provides a significant advancement in multispectral image stitching.