A Grid-Based Method for Removing Overlaps of Dimensionality Reduction Scatterplot Layouts

Area of Science:

• Data Visualization
• Information Visualization
• Computer Graphics

Background:

• Dimensionality Reduction (DR) scatterplots are widely used for multidimensional data analysis.
• Occlusion in DR scatterplots, especially with informative glyphs, hinders data interpretation.
• Existing overlap removal methods often compromise layout integrity or glyph visibility.

Purpose of the Study:

• To introduce Distance Grid (DGrid), a novel post-processing strategy for overlap removal in DR scatterplots.
• To address the limitations of current methods that distort layouts or reduce glyph sizes.
• To preserve original scatterplot characteristics and ensure glyph readability.

Main Methods:

• Developed Distance Grid (DGrid) as a post-processing technique for DR scatterplot layouts.
• Implemented DGrid to remove glyph overlaps while maintaining layout fidelity.
• Conducted extensive comparative evaluations and a user study to assess performance.

Main Results:

• DGrid significantly outperforms state-of-the-art methods in overlap removal across multiple metrics.
• DGrid demonstrates superior speed, particularly for large datasets.
• User studies confirm DGrid's effectiveness in preserving visual characteristics and aesthetics.

Conclusions:

• DGrid offers an effective and efficient solution for overlap removal in DR scatterplots.
• The method successfully balances overlap elimination with the preservation of data structure and visual quality.
• DGrid represents a significant advancement in the visualization of high-dimensional data.