398 Downloads (Pure)

Abstract

We present a method for supervised volumetric segmentation
based on a dictionary of small cubes composed of pairs of intensity
and label cubes. Intensity cubes are small image volumes where each
voxel contains an image intensity. Label cubes are volumes with voxelwise
probabilities for a given label. The segmentation process is done
by matching a cube from the volume, of the same size as the dictionary
intensity cubes, to the most similar intensity dictionary cube, and from
the associated label cube we get voxel-wise label probabilities. Probabilities
from overlapping cubes are averaged and hereby we obtain a robust
label probability encoding. The dictionary is computed from labeled volumetric
image data based on weighted clustering. We experimentally
demonstrate our method using two data sets from material science – a
phantom data set of a solid oxide fuel cell simulation for detecting three
phases and their interfaces, and a tomogram of a glass fiber composite
used in wind turbine blades for detecting individual glass fibers.
Original languageEnglish
Title of host publicationImage Analysis : 19th Scandinavian Conference, SCIA 2015 Copenhagen, Denmark, June 15–17, 2015 Proceedings
Number of pages12
PublisherSpringer Science+Business Media
Publication date2015
Pages504-515
ISBN (Print)978-3-319-19664-0
ISBN (Electronic)978-3-319-19665-7
DOIs
Publication statusPublished - 2015
Event19th Scandinavian Conference on Image Analysis - Copenhagen, Denmark
Duration: 15 Jun 201517 Jun 2015
Conference number: 19
http://www.scia2015.org/

Conference

Conference19th Scandinavian Conference on Image Analysis
Number19
CountryDenmark
CityCopenhagen
Period15/06/201517/06/2015
Internet address
SeriesLecture Notes in Computer Science
ISSN0302-9743

Keywords

  • Volume segmentation
  • Materials images
  • X-ray tomography
  • Learning dictionaries
  • Glass fiber segmentation

Fingerprint Dive into the research topics of 'Dictionary Based Segmentation in Volumes'. Together they form a unique fingerprint.

Cite this