Monica Jane Emerson: My research concerns segmentation of volumes that have been acquired through x-ray computed tomography. These volumes can be scans from various materials used in the energy sector, for example fibre composites or fuel cells.
I am working with a supervised segmentation method that is based on a so-called
dictionary of image patches and corresponding label patches (Dahl & Larsen, 2011).
This dictionary aims to represent in a compact manner all the information contained
in a set of training images that have been created through manual annotation. When
an image has to be segmented, instead of looking up the whole set of training
images, the dictionary is searched.
I have performed segmentations of solid oxide fuel cells and glass fibres uding 3D
methods (Emerson, Jespersen, Larsen, & Dahl, 2015), but at the moment I am
mainly performing the segmentations in 2D.
Currently I am working on the characterisation of uni-directional glass and carbon
fibre reinforced polymers. I have been able to segment the individual fibres from
scans with low quality and high fibre volume fraction. This involved segmentation of
the centre points of each fibre and a subsequent tracking procedure so as to connect
the points that belong to the same fibre, see Figure 1. This opens up for different
possibilities regarding the study of fibre architecture, such as determining the fibre
orientation distribution, which affects the compression strength of the material.
Future work will focus on modeling the fibres statistically, so as to for example detect
cracks. The aim is also to analyse these materials at a coarser scale where individual
fibres cannot be distinguished, but instead complete fibre bundles can be
characterised.
READ MORE: Figures and Bibliography (pdf)
Effective start/end date 01/09/2014 → 30/09/2018
Published as PhD report: Statistical Image Analysis of Tomograms with Application to Fibre Geometry Characterisation
Supervisors: Anders Bjorholm Dahl, Knut Conradsen
Section for Visual Computing