Robust point matching via vector field consensus.

Area of Science:

• Computer Vision
• Computational Geometry
• Machine Learning

Background:

• Establishing accurate point correspondences is crucial for many computer vision tasks.
• Existing methods often struggle with a high percentage of outliers in putative correspondences.

Purpose of the Study:

• To propose an efficient and robust algorithm for establishing point correspondences.
• To handle a large number of outliers in point matching problems.

Main Methods:

• Vector Field Consensus algorithm based on maximum a posteriori (MAP) estimation.
• Utilizes a Bayesian model with latent variables and the EM algorithm.
• Employs nonparametric geometrical constraints via Tikhonov regularizers in a reproducing kernel Hilbert space.

Main Results:

• Demonstrates robustness to up to 90% outliers in 2D and 3D datasets.
• Outperforms standard methods like RANSAC when a large number of outliers are present.
• Achieves computationally efficient and fast estimations, avoiding local minima.

Conclusions:

• Vector Field Consensus offers a robust and efficient solution for point correspondence problems with significant outliers.
• The nonparametric approach can be combined with parametric methods for enhanced geometric parameter estimation.
• The algorithm's generality allows application to other domains, including learning from corrupted data.