@TechReport{Cech-TR-2007-10,
  IS = { zkontrolovano 08 Aug 2014 },
  UPDATE  = { 2014-07-25 },
  author =       {{\v C}ech, Jan and {\v S}{\' a}ra, Radim},
  title =        {Windowpane Detection based on Maximum Aposteriori Labeling},
  institution =  {Center for Machine Perception, K13133 FEE Czech Technical
                  University},
  address =      {Prague, Czech Republic},
  year =         {2007},
  month =        {April},
  type =         {Research Report},
  number =       {CTU--CMP--2007--10},
  issn =         {1213-2365},
  pages =        {12},
  figures =      {7},
  authorship =   {50-50},
  project =      {FP6-IST-027113 eTRIMS},
  annote =       {Segmentation of windowpanes in the images of facades is
                  formulated as a task of maximum aposteriori
                  labeling. Assuming orthographic rectification of the
                  building facade, the windowpanes are always axis-parallel
                  rectangles of relatively low variability in
                  appearance. Every image pixel has one of 10 possible labels,
                  and the adjacent pixels are interconnected via links which
                  defines allowed label configuration, such that the labels
                  are forced to form a set of non-overlapping rectangles. The
                  task of finding the most probable labeling of a given image
                  leads to NP-hard discrete optimization problem. However, we
                  find an approximate solution using a general solver suitable
                  for such problems and we obtain promising results which we
                  demonstrate on several experiments. Substantial difference
                  between the presented paper and state-of-the-art papers on
                  segmentation based on Markov Random Fields is that we have a
                  strong structure model, forcing the labels to form
                  rectangles, while other methods does not model the structure
                  at all, they typically only have a penalty when adjacent
                  labels are different, in order to make resulting patches
                  more continuous to reduce influence of noise and prevent
                  over-segmentation.},
  keywords =     {Segmentation, structure model, Markov Random Fields, MRF,
                  Potts model, labeling},
  comment =      {eTRIMS deliverable report},
}