Homography Propagation and Optimization for Wide-Baseline Street Image Interpolation

Area of Science:

• Computer Vision
• Image Processing
• Geometric Computer Vision

Background:

• Wide-baseline street image interpolation is a challenging problem.
• Current methods often require computationally expensive 3D reconstruction or deep neural networks.
• There is a need for efficient and lightweight interpolation techniques.

Purpose of the Study:

• To develop a lightweight and efficient method for wide-baseline street image interpolation.
• To improve the accuracy and robustness of image interpolation compared to existing methods.
• To demonstrate that homography computation alone is sufficient for effective interpolation.

Main Methods:

• Estimating piecewise smooth homographies using simple homography fitting and refining operators.
• Combining homography fitting and propagation based on superpixel reliability.
• Integrating homography and mesh warping with a novel homography-constrained warping formulation for smoothness.

Main Results:

• The proposed method significantly increases the accuracy and robustness of estimated homographies.
• The homography-constrained warping formulation effectively eliminates artifacts like overlapping and stretching.
• The method achieves state-of-the-art performance in wide-baseline image interpolation.
• Experiments on diverse datasets validate the efficiency and effectiveness.

Conclusions:

• Homography computation is sufficient for achieving high-quality wide-baseline image interpolation.
• The proposed lightweight method offers a practical and efficient solution for street image interpolation.
• The technique improves upon existing approaches in terms of accuracy, robustness, and computational cost.