A Graph-Based Superpixel Segmentation Approach Applied to Pansharpening

Area of Science:

• Remote Sensing
• Image Processing
• Computer Vision

Background:

• Pansharpening fuses high-resolution panchromatic (PAN) and low-resolution multispectral (MS) images to create a high-resolution MS image.
• Existing methods' performance depends heavily on the fusion strategy and gain coefficient estimation.
• Graph-based segmentation offers a promising approach for guiding spatial information injection.

Purpose of the Study:

• To propose an image-driven regional pansharpening technique using simplex optimization and graph-based superpixel segmentation.
• To compute injection gains over a graph-driven segmentation map for optimal fusion.
• To generate a comprehensive high-resolution MS image with enhanced spatial and spectral information.

Main Methods:

• Utilizing simplex optimization analysis for image fusion.
• Applying Simple Linear Iterative Clustering (SLIC) for superpixel segmentation on MS images.
• Constructing a Region Adjacency Graph (RAG) for segmentation map merging.
• Employing a graph-driven segmentation map to guide spatial information injection.
• Inferring local details using a local simplex injection fusion rule.

Main Results:

• The proposed method optimally combines spatial and spectral information.
• Graph-based segmentation guides the injection of spatial details effectively.
• The technique generates a unique, comprehensive high-resolution MS image.
• Quality improvements were validated on GeoEye-1 and WorldView-3 datasets at reduced and full scales.

Conclusions:

• The proposed pansharpening technique effectively enhances MS image resolution.
• Graph-based segmentation provides a robust framework for fusion parameter optimization.
• The method demonstrates significant quality improvements in generated high-resolution MS images.