||Estimation of fat layer thickness profiles in carcass midsections
||Skytte, Jacob Lercke (Image Analysis and Computer Graphics, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
||Ersbøll, Bjarne Kjær (DTU Data Analysis, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
||Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
||This thesis describes a method for automatically extracting fat layer thickness
profiles from CT scanned pig carcass midsections. A fat layer thickness profile
can be constructed as shortest distance correspondences between the outlines of
skin and muscle, both of which can be segmented out separately.
The image data is preprocessed before further use. A virtual tripartition is carried
out in order to isolate the carcass midsection. Irrelevant voxels and objects
are removed using connected component analysis. Further voxel anomalies are
found and repaired using blob detection and a simple reconstruction scheme.
Segmentation of the skin outline is carried out using an ad hoc method, which
utilizes mathematical morphology, edge tracing, and polar transformation.
The muscle outline is segmented out, by trying the following three-dimensional
classification schemes: anisotropic diffusion, 3D contextual Bayesian classification,
and Markov random field segmentation. These methods are also held up
against a two-dimensional classification method, in order to compare the stability
towards noise in this type of image data.
The intensity range for skin and muscle voxels overlaps, and therefore skin is
included in the final muscle classification. This skin is afterwards removed using
the, already, segmented skin outline along with distance maps.
As the segmented outlines reside in the discrete domain, a guided smoothing
is applied using active contours. Thus, a more biological behaviour is obtained
as well as sub-voxel accuracy. From the smooth outlines, the final fat layer
thickness profiles are created.
Finally potential applications, for the fat layer thickness profiles, are presented
in both theory and practice.
Due to the lack of ground truth, all results are subject to evaluation by visual
inspection, which reveals overall good results for the provided image data.
||Technical University of Denmark (DTU) : Kgs. Lyngby, Denmark
Creation date: 2010-06-03
Update date: 2010-12-16