@InProceedings{BPH13,
  author =      {Bresler, Martin and Pr{\r u}{\v s}a, Daniel and Hlav{\'a}{\v c}, V{\' a}clav},
  title =       {Simultaneous Segmentation and Recognition of Graphical Symbols using a Composite Descriptor},
  c_title =     {Soub{\v e}{\v z}n{\' a} segmentace a rozpozn{\'a}v{\'a}n{\'i} grafick{\'y}ch 
                 symbol{\r u} pomoc{\' i} kompozitn{\'i}ho deskriptoru},
  year =        {2013},
  pages =       {16-23},
  booktitle =   {CVWW 2013: Proceedings of the 18th Computer Vision Winter Workshop},
  publisher =   {Vienna University of Technology},
  address =     {Karlsplatz 13, Vienna, Austria},
  editor =      {Kropatsch, Walter G. and Ramachandran, Geetha and Torres, Fuensanta},
  book_pages =  {125},
  isbn =        {978-3-200-02943-9},
  month =       {February},
  day =         {4-6},
  venue =       {Hernstein, Austria},
  annote =      {This work deals with recognition of hand-drawn
    graphical symbols in diagrams. We present two
    contributions. First, we designed a new composite descriptor
    expressing overall appearance of symbols. We achieved rather
    favorable accuracy in classification of segmented symbols on
    benchmark databases, which is 98.93 prec. for a database of flow
    charts, 98.33 prec. for a database of crisis management icons, and
    92.94 perc. for a database of digits. Second, we used the
    descriptor in the task of simultaneous segmentation and
    recognition of graphical symbols. Our method creates symbol
    candidates by grouping spatially close strokes. Symbol candidates
    are classified by a multiclass SVM classifier learned on a dataset
    with negative examples. Thus, some portion of the candidates is
    filtered out. The joint segmentation and classification was tested
    on diagrams from the flowchart database. We were able to find
    91.85 prec. of symbols while generating 8.8 times more symbol
    candidates than is the number of true symbols per diagram in
    average.},
  keywords =    {Diagram recognition, Flowchart2076-1465s, Pattern recognition, SVM},
  prestige =    {international},
  project =     {GACR P103/10/0783},
  status =      {published},
}