Advanced Measurement and Quantification of Industrial CT Data

Fritz-Michael Gschwantner
Advanced Measurement and Quantification of Industrial CT Data
[ Thesis]
Content:

Information

Abstract

Non–destructive testing (NDT) is a vital part in today’s industrial production and research processes. Such testing procedures often use Computed Tomography (CT) in order to get insights of the inner parts of an object. During the analysis of different objects, certain features can be automatically segmented and quantified in the CT dataset. However, due to various effects during the acquisition of the data, the original boundaries of two materials within the objects are not accurately represented in the dataset. This thesis describes a method to reconstruct these boundaries for automatically segmented features on a subvoxel level of the dataset. They are searched along the gradient of the data, using an edge–detection approach commonly used in image processing. The result is then represented as a distance field and further quantified through over–sampling and measuring. For a variety of datasets it is shown that these reconstructed boundaries are indeed providing a more accurate representation of the original segmented region. Further comparisons are made with a method that simply tries to improve the visual appearance through smoothing.

Additional Files and Images

Additional files:
Thesis
Thesis




BibTeX

Download BibTeX-Entry
@mastersthesis{Gschwantner_2011_AMQ,
  title =      "Advanced Measurement and Quantification of Industrial CT
               Data",
  author =     "Fritz-Michael Gschwantner",
  year =       "2011",
  abstract =   "Non–destructive testing (NDT) is a vital part in
               today’s industrial production and research processes.
               Such testing procedures often use Computed Tomography (CT)
               in order to get insights of the inner parts of an object.
               During the analysis of different objects, certain features
               can be automatically segmented and quantified in the CT
               dataset. However, due to various effects during the
               acquisition of the data, the original boundaries of two
               materials within the objects are not accurately represented
               in the dataset. This thesis describes a method to
               reconstruct these boundaries for automatically segmented
               features on a subvoxel level of the dataset. They are
               searched along the gradient of the data, using an
               edge–detection approach commonly used in image
               processing. The result is then represented as a distance
               field and further quantified through over–sampling
               and measuring. For a variety of datasets it is shown that
               these reconstructed boundaries are indeed providing a more
               accurate representation of the original segmented region.
               Further comparisons are made with a method that simply tries
               to improve the visual appearance through smoothing.",
  address =    "Favoritenstrasse 9-11/186, A-1040 Vienna, Austria",
  school =     "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology",
  month =      oct,
  URL =        "http://www.cg.tuwien.ac.at/research/publications/2011/Gschwantner_2011_AMQ/",
}