IEEE TVCG 2026

Distortion-aware brushing

An interaction technique for more reliable cluster analysis in multidimensional projections by continuously relocating points during hover and brushing.

Published as Distortion-Aware Brushing for Reliable Cluster Analysis in Multidimensional Projections, IEEE Transactions on Visualization and Computer Graphics, 32(2), 2026, pp. 2165-2182. The paper evaluates the technique through controlled studies and practical use cases. DOI / PDF

Open Demo Read Paper Browse Repositories

Demo preview Static preview

Open demo in new tab

Why

Why brushing on projections is hard

A multidimensional projection compresses high-dimensional structure into a 2D layout, and that layout can be misleading for cluster analysis. Points that look nearby on screen may be far apart in the original data space, while true neighbors in the high-dimensional space may appear split across the projection.

Distortion-aware brushing addresses that reliability problem during interaction itself. Instead of treating the current 2D arrangement as fixed truth, it continuously relocates points around the brushed set so the visible cluster better reflects the underlying multidimensional structure.

In paper terms, the main failure modes are missing neighbors and false neighbors. Distortion-aware brushing handles both through continuous relocation during hover and brushing.

Repositories

Project repositories

The project is split into small repositories so preprocessing, runtime logic, web integration, and the public demo can evolve independently.

@dabrush/web Integrated Canvas and React web library. This is the main entry point for most users.
@dabrush/preprocess-js Browser-friendly preprocessing for aligned HD and LD data.
dabrush-preprocess Python package and CLI for heavier preprocessing workloads.
@dabrush/engine DOM-free brushing state machine for custom renderers and integrations.
@dabrush/schema Shared dataset contract and runtime validation.
dab-demo 300-point Fashion-MNIST demo deployed on GitHub Pages.

Organization: github.com/orgs/distortion-aware-brushing/repositories

BibTeX

Citation

Download citation.bib

@ARTICLE{11184260,
  author={Jeon, Hyeon and Aupetit, Michaël and Lee, Soohyun and Ko, Kwon and Kim, Youngtaek and Quadri, Ghulam Jilani and Seo, Jinwook},
  journal={IEEE Transactions on Visualization and Computer Graphics},
  title={Distortion-Aware Brushing for Reliable Cluster Analysis in Multidimensional Projections},
  year={2026},
  volume={32},
  number={2},
  pages={2165-2182},
  keywords={Distortion;Brushes;Shape;Reliability;Layout;Data visualization;Visual analytics;Nonlinear distortion;Labeling;Data mining;Multidimensional projections (MDPs);distortion-aware brushing;brushing;distortions;visual clustering;cluster analysis},
  doi={10.1109/TVCG.2025.3615314}}