Publication detail

Segmentation of multi-phase object applying trainable segmentation

KALASOVÁ, D. MAŠEK, J. ZIKMUND, T. SPURNÝ, P. HALODA, J. BURGET, R. KAISER, J.

Original Title

Segmentation of multi-phase object applying trainable segmentation

English Title

Segmentation of multi-phase object applying trainable segmentation

Type

journal article - other

Language

en

Original Abstract

In X-ray computed tomography (CT), post-processing of acquired data is necessary for obtaining quantitative information of the object. As initial step, it is necessary to segment different materials of the sample. The easiest and standardly used segmentation method is based on global thresholding according to histogram, but it works well only if histogram with multi-modal character where the intensity is distributed to the separate count peaks. In this paper, we show the possibility of segmentation of tomographic data using trainable segmentation on data, where standard global thresholding fails. Trainable segmentation is a method that combines a collection of machine learning algorithms (decision tree, neural network, etc.) with a set of selected image features to produce binary pixel-based segmentation. This method is demonstrated on a sample of meteorite consisting of multiple phases (silicates, metals, sulphides), where knowledge of volumes of different materials is important for non-destructive study of modal phase composition, meteorite microstructures and identification of lithologies with different origin and evolution.

English abstract

In X-ray computed tomography (CT), post-processing of acquired data is necessary for obtaining quantitative information of the object. As initial step, it is necessary to segment different materials of the sample. The easiest and standardly used segmentation method is based on global thresholding according to histogram, but it works well only if histogram with multi-modal character where the intensity is distributed to the separate count peaks. In this paper, we show the possibility of segmentation of tomographic data using trainable segmentation on data, where standard global thresholding fails. Trainable segmentation is a method that combines a collection of machine learning algorithms (decision tree, neural network, etc.) with a set of selected image features to produce binary pixel-based segmentation. This method is demonstrated on a sample of meteorite consisting of multiple phases (silicates, metals, sulphides), where knowledge of volumes of different materials is important for non-destructive study of modal phase composition, meteorite microstructures and identification of lithologies with different origin and evolution.

Keywords

segmentation, trainable segmentation, machine learning, image processing

Released

09.02.2017

Publisher

NDT.net

Pages from

1

Pages to

6

Pages count

6

URL

BibTex


@article{BUT133386,
  author="Dominika {Kalasová} and Jan {Mašek} and Tomáš {Zikmund} and Pavel {Spurný} and Jakub {Haloda} and Radim {Burget} and Jozef {Kaiser}",
  title="Segmentation of multi-phase object applying trainable segmentation",
  annote="In X-ray computed tomography (CT), post-processing of acquired data is necessary for obtaining quantitative information of the object. As initial step, it is necessary to segment different materials of the sample. The easiest and standardly used segmentation method is based on global thresholding according to histogram, but it works well only if histogram with multi-modal character where the intensity is distributed to the separate count peaks.
In this paper, we show the possibility of segmentation of tomographic data using trainable segmentation on data, where standard global thresholding fails. Trainable segmentation is a method that combines a collection of machine learning algorithms (decision tree, neural network, etc.) with a set of selected image features to produce binary pixel-based segmentation. This method is demonstrated on a sample of meteorite consisting of multiple phases (silicates, metals, sulphides), where knowledge of volumes of different materials is important for non-destructive study of modal phase composition, meteorite microstructures and identification of lithologies with different origin and evolution.",
  address="NDT.net",
  booktitle="Conference proceedings: 7th Conference on Industrial Computed Tomography (ICT) 2017",
  chapter="133386",
  howpublished="online",
  institution="NDT.net",
  number="2017",
  year="2017",
  month="february",
  pages="1--6",
  publisher="NDT.net",
  type="journal article - other"
}