Automatic Model Generation for Sparse MRF Appearance Models using Minimum Description Length (bibtex)
by Eva Dittrich
Abstract:
This report explores the combination of the Minimum Description Length (MDL) approach and Sparse Markov Random Fields Appearance Models (SAMs). SAMs are a method to locate a structure that is learnt from annotated training data in a new and unseen image. However, to achieve this result it is necessary to provide the SAMs with manual annotations of the images (landmarks) for a large set of training examples, which is a time consuming and error-prone requirement. The goal of this work is to become independent from manual annotations and to obtain the annotations automatically by using an MDL based approach. We report experimental results for different types of data (synthetic data, 2D X-rays and 3D CTs) and the method was modified to reach the best possible results for each of them. The resulting approach allows to construct SAMs in a fully automated fashion.
Reference:
Automatic Model Generation for Sparse MRF Appearance Models using Minimum Description Length (Eva Dittrich), Technical report, PRIP, TU Wien, 2009.
Bibtex Entry:
@TechReport{TR120,
  author =	 "Eva Dittrich",
  title =	 "Automatic Model Generation for Sparse MRF Appearance Models using Minimum Description Length",
  institution =	 "PRIP, TU Wien",
  number =	 "PRIP-TR-120",
  year =	 "2009",
  url =		 "ftp://ftp.prip.tuwien.ac.at/pub/publications/trs/tr120.pdf",
  abstract =	 "This report explores the combination of the Minimum Description Length (MDL) approach
and Sparse Markov Random Fields Appearance Models (SAMs). SAMs are a method to
locate a structure that is learnt from annotated training data in a new and unseen image.
However, to achieve this result it is necessary to provide the SAMs with manual annotations
of the images (landmarks) for a large set of training examples, which is a time consuming
and error-prone requirement. The goal of this work is to become independent from manual
annotations and to obtain the annotations automatically by using an MDL based approach.
We report experimental results for different types of data (synthetic data, 2D X-rays and
3D CTs) and the method was modified to reach the best possible results for each of them.
The resulting approach allows to construct SAMs in a fully automated fashion.",
}
Powered by bibtexbrowser