Searching for Representative Modes on Hypergraphs for Robust Geometric Model Fitting.

Area of Science:

• Computer Vision
• Machine Learning
• Geometric Modeling

Background:

• Geometric model fitting is crucial for data segmentation.
• Existing methods struggle with severe outliers and complex data structures.
• Hypergraph-based approaches offer potential for capturing higher-order relationships.

Purpose of the Study:

• To propose a robust geometric model fitting method for multi-structure data.
• To address challenges posed by severe outliers in model fitting and segmentation.
• To develop an efficient algorithm for identifying model parameters and counts.

Main Methods:

• Constructing a hypergraph where vertices represent model hypotheses and hyperedges represent data points.
• Utilizing higher-order similarities within the hypergraph to capture complex relationships.
• Developing a hypergraph reduction technique to simplify the model.
• Proposing a novel mode-seeking algorithm based on vertex weighting and similarity analysis.

Main Results:

• The method effectively fits and segments multi-structure data in the presence of severe outliers.
• The hypergraph reduction and mode-seeking algorithm efficiently process complex data.
• The proposed method demonstrates significant superiority over state-of-the-art techniques on synthetic and real data.

Conclusions:

• The proposed hypergraph-based geometric model fitting method is efficient and effective.
• It accurately estimates the number and parameters of model instances simultaneously.
• This approach offers a robust solution for challenging data segmentation tasks.