@TechReport{Petricek-TR-2010-09,
  IS = { zkontrolovano 31 Jan 2011 },
  UPDATE  = { 2010-09-16 },
   author =       {Pet{\v r}{\'\i}{\v c}ek, Tom{\'a}{\v s} and 
                   Svoboda, Tom{\'a}{\v s}},
   title =        {Matching by Normalized Cross-Correlation---Reimplementation,
                   Comparison to Invariant Features},
   institution =  {Center for Machine Perception, 
                   K13133 FEE Czech Technical University},
   address =      {Prague, Czech Republic},
   year =         {2010},
   month =        {July},
   type =         {Research Report},
   number =       {CTU--CMP--2010--09},
   issn =         {1213-2365},
   pages =        {24},
   authorship =   {50-50},
   psurl =        {[Petricek-TR-2010-09.pdf]},
   project =      {GACR P103/10/1585, FP7-ICT-247870 NIFTi},
   annote =       {The normalized cross-correlation is one of the most
     popular methods for image matching. While fast implementations of
     the algorithm are available in standard mathematical toolboxes,
     there still are ways to get significant speed-up for many
     practical applications. This work investigates the following
     possibilities: reusing image sums for matching multiple
     templates, using maximum expected disparity to bound search
     regions, and using downscaling factor to reduce size of
     computation. Based on our experiments we conclude that both
     downscaling images and bounding disparity field yields
     significant speed-up. Downscaling images also yields higher
     repeatability rate, which remains reasonably high for downscaling
     factors up to 5. For images related by translation, matching by
     normalized cross-correlation gives higher repatability rate and
     matching score than invariant features with SIFT decriptors.},
   keywords =     {computer vision, image matching},
}