IEEE VIS 2024 Content: Uniform Sample Distribution in Scatterplots via Sector-based Transformation

Uniform Sample Distribution in Scatterplots via Sector-based Transformation



Hennes Rave - University of Münster, Münster, Germany

 Vladimir Molchanov - University of Münster, Münster, Germany

 Lars Linsen - University of Münster, Münster, Germany

 Room: Bayshore VI

2024-10-16T13:15:00ZGMT-0600Change your timezone on the schedule page
2024-10-16T13:15:00Z

Exemplar figure, described by caption below — Sector-based transformation of a UMAP embedding of the Iris dataset. 16 sectors and anchor points for a selected sample are shown for the original scatterplot. The black anchor point at the bottom belongs to the highlighted sector at the top. Samples are moved toward a sector's anchor point based on the point density inside that sector. The resulting displacement vector is shown in blue.

Fast forward

Full Video

Keywords

Scatterplot de-cluttering, spatial transformation.

Abstract

A high number of samples often leads to occlusion in scatterplots, which hinders data perception and analysis. De-cluttering approaches based on spatial transformation reduce visual clutter by remapping samples using the entire available scatterplot domain. Such regularized scatterplots may still be used for data analysis tasks, if the spatial transformation is smooth and preserves the original neighborhood relations of samples. Recently, Rave et al. proposed an efficient regularization method based on integral images. We propose a generalization of their regularization scheme using sector-based transformations with the aim of increasing sample uniformity of the resulting scatterplot. We document the improvement of our approach using various uniformity measures.