Applying manifold learning to plotting approximate contour trees.

Area of Science:

• Topological Data Analysis
• Computer Graphics
• Scientific Visualization

Background:

• Contour trees are vital for analyzing isosurface topology and temporal changes in volumetric data.
• Existing contour tree algorithms can be complex, particularly for high-dimensional datasets like time-varying volumes.

Purpose of the Study:

• To develop a simpler, effective method for generating approximate contour trees from scattered high-dimensional data.
• To enable efficient feature extraction and analysis of volumes and their temporal evolution.

Main Methods:

• Utilizing manifold learning to embed high-dimensional data into a lower-dimensional space while preserving local structure.
• Introducing novel distance metrics for manifold learning to adapt existing dimensionality reduction techniques.
• Implementing data size reduction and segmentation for coarse-to-fine analysis of large datasets.

Main Results:

• Demonstrated a novel approach to plotting approximate contour trees in 3D space from scattered samples.
• Successfully adapted manifold learning with new distance metrics for contour tree construction.
• Enabled efficient analysis of volumetric features and temporal behaviors through constructed contour trees.

Conclusions:

• The proposed manifold learning-based approach offers a simplified and effective method for contour tree generation.
• This technique facilitates the analysis of complex, high-dimensional data, including time-varying volumes.
• The approach provides a robust tool for feature traversal and understanding temporal dynamics in datasets.