Fast parameter prediction for Active Appearance Models using Canonical Correlation Analysis - An AAM Matlab Implementation (bibtex)
by Ren�Donner, Georg Langs, Michael Reiter, Horst Bischof, Robert Sablatnig
Abstract:
Active Appearance Models (AAM) provide a compact statistical model of data encom- passing both shape and texture variations. This report introduces a novel and fast search algorithm for AAMs based on canonical correlation analysis (CCA). In contrast to the standard AAM matching approach CCA exploits the correlation between texture residuals and model parameters more efficiently. In a set of experiments using face and medical images we show that CCA based search consistently outperforms the convergence speed of the standard method by a factor of four. The time needed for training is reduced by 80%. Since the implementation effort of the standard approach and CCA are similar our results suggest that CCA can replace the standard AAM search.
Reference:
Fast parameter prediction for Active Appearance Models using Canonical Correlation Analysis - An AAM Matlab Implementation (Ren�Donner, Georg Langs, Michael Reiter, Horst Bischof, Robert Sablatnig), Technical report, PRIP, TU Wien, 2005.
Bibtex Entry:
@TechReport{TR097,
  author =	 "Ren�Donner and Georg Langs and Michael Reiter and
                  Horst Bischof and Robert Sablatnig",
  title =	 "Fast parameter prediction for Active Appearance
                  Models using Canonical Correlation Analysis - An AAM
                  Matlab Implementation",
  institution =	 "PRIP, TU Wien",
  number =	 "PRIP-TR-097",
  year =	 "2005",
  url =		 "ftp://ftp.prip.tuwien.ac.at/pub/publications/trs/tr97.pdf",
  abstract =	 "Active Appearance Models (AAM) provide a compact
                  statistical model of data encom- passing both shape
                  and texture variations. This report introduces a
                  novel and fast search algorithm for AAMs based on
                  canonical correlation analysis (CCA). In contrast to
                  the standard AAM matching approach CCA exploits the
                  correlation between texture residuals and model
                  parameters more efficiently. In a set of experiments
                  using face and medical images we show that CCA based
                  search consistently outperforms the convergence
                  speed of the standard method by a factor of
                  four. The time needed for training is reduced by
                  80%. Since the implementation effort of the standard
                  approach and CCA are similar our results suggest
                  that CCA can replace the standard AAM search.",
}
Powered by bibtexbrowser