Multiple Manifold Clustering Using Curvature Constrained Path.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Multiple surface clustering is complex, particularly with intersecting surfaces.
• Existing methods like Isomap struggle with accuracy near intersections due to shortest path limitations.
• Isomap's shortest path algorithm lacks curvature constraints, allowing paths between distinct surfaces.

Purpose of the Study:

• To address the limitations of current surface clustering methods for intersecting surfaces.
• To develop an improved clustering algorithm by incorporating curvature constraints into Isomap.
• To enhance the accuracy of multiple surface clustering, especially in complex geometric scenarios.

Main Methods:

• Imposed a curvature constraint on the shortest path algorithm within the Isomap framework.
• Selected random landmark nodes and identified curvature-constrained paths between them and other nodes.
• Constructed binary feature vectors representing point connectivity to landmarks for clustering input.

Main Results:

• The proposed method demonstrates improved performance in multiple surface clustering tasks.
• It effectively handles intersecting surfaces where traditional methods fail.
• Performance is comparable to state-of-the-art techniques like K-manifold and spectral multi-manifold clustering.

Conclusions:

• The curvature-constrained shortest path approach significantly enhances multiple surface clustering accuracy.
• This method offers a robust solution for clustering complex, intersecting surfaces.
• The approach provides a viable alternative to existing advanced clustering algorithms.