Smoothness-Driven Consensus Based on Compact Representation for Robust Feature Matching.

Area of Science:

• Computer Vision
• Machine Learning
• Data Analysis

Background:

• Robust feature matching relies on distinguishing true correspondences (inliers) from false ones (outliers) by recovering smooth functions from data.
• Existing regularization methods offer theoretical advantages but suffer from high time and space complexities, hindering practical application.

Purpose of the Study:

• To propose a novel, computationally efficient method for multivariate regression and point matching.
• To exploit the sparsity structure of smooth functions for improved feature matching scalability and robustness.

Main Methods:

• Utilizes compact Fourier bases for function construction, enabling a coarse-to-fine representation and explicit imposition of smoothness constraints.
• Formulates the learning problem within a Bayesian framework using latent variables and a mixture model to handle gross outliers.
• Derives a fast expectation-maximization algorithm for efficient parameter estimation.

Main Results:

• The proposed method demonstrates reduced computational complexities compared to traditional regularization techniques.
• Achieves superior performance in terms of scalability and robustness on synthetic data, image matching, and point set registration tasks.
• Outperforms current state-of-the-art methods in extensive experimental evaluations.

Conclusions:

• The novel approach effectively addresses the limitations of existing methods for robust feature matching.
• Offers a scalable and robust solution for multivariate regression and point matching problems, particularly in the presence of outliers.
• The exploitation of sparse structures in smooth functions provides significant advantages for real-world applications.