Exploring Uncertainty in Image Segmentation Ensembles

Bernhard Fröhler, Torsten Möller, Johannes Weissenböck, Hans-Christian Hege, Johann Kastner, Christoph Heinzl

Publikation: Beitrag in Buch/Bericht/TagungsbandKonferenzbeitrag

Abstract

Finding the most accurate image segmentation involves analyzing results from different algorithms or parameterizations. In this work, we identify different types of uncertainty in this analysis that are represented by the results of probabilistic algorithms, by the local variability in the segmentation, and by the variability across the segmentation ensemble. We propose visualization techniques for the analysis of such types of uncertainties in segmentation ensembles. For a global analysis we provide overview visualizations in the image domain as well as in the label space. Our probability probing and scatter plot based techniques facilitate a local analysis. We evaluate our techniques using a case study on industrial computed tomography data.
Titel in ÜbersetzungExploring Uncertainty in Image Segmentation Ensembles
OriginalspracheDeutsch
TitelProceedings of the 20th EG/VGTC Conference on Visualization
Seiten33-35
Seitenumfang3
DOIs
PublikationsstatusVeröffentlicht - 2018
VeranstaltungEurographics Conference on Visualization 2018 (EuroVis 2018) - Brünn, Tschechische Republik
Dauer: 4 Juni 20188 Juni 2018
https://www.eurovis2018.org/

Konferenz

KonferenzEurographics Conference on Visualization 2018 (EuroVis 2018)
Land/GebietTschechische Republik
OrtBrünn
Zeitraum04.06.201808.06.2018
Internetadresse

Fingerprint

Untersuchen Sie die Forschungsthemen von „Exploring Uncertainty in Image Segmentation Ensembles“. Zusammen bilden sie einen einzigartigen Fingerprint.

Zitieren