Coupled Shape Model Segmentation in Pig Carcasses |
|
| Abstract | In this paper we are concerned with multi-object segmentation.
For each object we will train a level set function
based shape prior from a sample set of outlines. The outlines
are aligned in a multi-resolution scheme wrt. an Euclidean
similarity transformation in order to maximize the
overlap of the interior between all pairs of outlines. Then
the outlines are converted to level set functions. A shape
model is constructed from the mean level set and the first
few principal variations. We combine the prior model with
an observation model based on the Chan-Vese functional
assuming constant intensity levels inside the outline as well
as in a narrow band outside the outline. The maximum a
posteriori estimate of the outline is found by gradient descent
optimization. In order to segment a group of mutually
dependent objects we propose 2 procedures, 1) the objects
are found sequentially by conditioning the initialization of
the next search from already found objects; 2) all objects
are found simultaneously and a repelling force is introduced
in order to avoid overlap between outlines in the solution.
The methods are applied to segmentation of cross sections
of muscles in slices of CT scans of pig backs for quality assessment of bacon slices. |
| Type | Conference paper [With referee] |
| Conference | ICPR 2006, IEEE International Conference on Pattern Recognition 2006 |
| Year | 2006 |
| Electronic version(s) | [pdf] |
| BibTeX data | [bibtex] |
| IMM Group(s) | Image Analysis & Computer Graphics |