A dense matching method for remote sensing images fused with CPS denoising

Area of Science:

• Photogrammetry and Remote Sensing
• Computer Vision
• Image Processing

Background:

• Dense matching is essential for accurate 3D reconstruction from remote sensing imagery.
• Existing methods often struggle with noise and blur, impacting accuracy and efficiency.
• The Semi-Global Matching (SGM) algorithm is a common but improvable approach for stereo matching.

Purpose of the Study:

• To enhance the accuracy and efficiency of dense matching for remote sensing images.
• To integrate the CPS image denoising algorithm with SGM for improved performance.
• To evaluate the effectiveness of the proposed method on benchmark and real-world satellite datasets.

Main Methods:

• Developed an enhanced dense matching method combining CPS denoising with SGM.
• Implemented preprocessing steps: image cropping and pixel coordinate transformation.
• Utilized an epipolar line model to construct an epipolar image for efficient matching.
• Incorporated a PSNR-based criterion for adaptive denoising using the CPS algorithm.

Main Results:

• Reduced the average mismatch rate by 13.1% compared to the SGBM algorithm.
• Increased the running speed by approximately 3 to 4 times.
• The CPS denoising component alone decreased the mismatch rate by an average of 8.97%.
• Validated on Middlebury 2021 datasets and World-View3 satellite stereo image pairs.

Conclusions:

• The proposed CPS-enhanced SGM algorithm effectively addresses challenges like image blur and noise in dense matching.
• The method significantly improves both the accuracy and operational efficiency of remote sensing 3D reconstruction.
• This approach offers a robust solution for high-quality 3D modeling from challenging remote sensing data.