Volume Puzzle: visual analysis of segmented volume data with multivariate attributes



Marco Agus, Amal Aboulhassan, Khaled Ahmed Lutf Al-Thelaya, Giovanni Pintore, Enrico Gobbetti, Corrado Cali', Jens Schneider

 View presentation:2022-10-20T19:09:00ZGMT-0600Change your timezone on the schedule page
2022-10-20T19:09:00Z

Exemplar figure, described by caption below — Inspired by word search puzzles, we present Volume Puzzle, a framework that allows practitioners to interactively and/or automatically reveal spatial patterns from segmented volumes with associated multivariate attributes. For speeding up spatial analysis, we propose an algorithm that computes attribute projection through dimensionality reduction, kernel density estimation, and topological analysis based on the Morse-Smale complex. The framework can be used for explorative analysis in various domains, like material science or neuroscience.

Prerecorded Talk

The live footage of the talk, including the Q&A, can be viewed on the session page, Scientific Visualization, Ensembles, and Accessibility.

Fast forward

Keywords

Segmented Volumes, Multivariate data, Color mapping, Dimensionality reduction

Abstract

A variety of application domains, including material science, neuroscience, and connectomics, commonly use segmented volume data for explorative visual analysis. In many cases, segmented objects are characterized by multivariate attributes expressing specific geometric or physical features. Objects with similar characteristics, determined by selected attribute configurations, can create peculiar spatial patterns, whose detection and study is of fundamental importance. This task is notoriously difficult, especially when the number of attributes per segment is large. In this work, we propose an interactive framework that combines a state-of-the-art direct volume renderer for categorical volumes with techniques for the analysis of the attribute space and for the automatic creation of 2D transfer function. We show, in particular, how dimensionality reduction, kernel-density estimation, and topological techniques such as Morse analysis combined with scatter and density plots allow the efficient design of two-dimensional color maps that highlight spatial patterns. The capabilities of our framework are demonstrated on synthetic and real-world data from several domains.