Homography Ranking Based on Multiple Groups of Point Correspondences

Area of Science:

• Computer Vision
• Image Processing
• Geometric Transformations

Background:

• Homography mapping corrects perspective distortion using point correspondences from known markers.
• Estimating homographies becomes challenging with multiple markers of unknown relative positions, leading to indeterminate correspondences.
• Current methods often yield isolated homographies without a clear method for selecting the optimal one for the entire image.

Purpose of the Study:

• To develop a robust method for ranking and selecting the best homography from multiple markers in scenarios with unknown relative positions.
• To enhance the accuracy of perspective distortion removal in images with multiple markers.
• To extend existing post-processing techniques for homography estimation.

Main Methods:

• Proposing a novel post-processing method to rank isolated homographies derived from distinct markers.
• Leveraging point correspondences and assuming markers differ by similarity transformation after rectification.
• Validating the method's robustness using a synthetic dataset.

Main Results:

• The proposed method effectively ranks isolated homographies, enabling the selection of the most accurate one.
• Demonstrated robustness in selecting the optimal homography for image reprojection.
• Achieved approximately a 60% relative improvement compared to random selection strategies using OpenCV's homography estimation.

Conclusions:

• The developed homography ranking method significantly improves the selection of the best homography for correcting perspective distortion.
• This approach offers a valuable extension to existing computer vision techniques for multi-marker scenarios.
• The method provides a reliable solution for indeterminate point correspondences in image rectification.