Page 260 -
P. 260

248     6 Structural Pattern Recognition

                              Strings are useful for representing concatenated structures, for instance, segment
                           joining  in  signal  descriptions.  Even  in  some  apparently  complex  relations  of
                           primitives, describing 2D and 3D structures, it is still possible to use strings, as we
                           will see in section 6.3.4.
                              Sometimes the ordered sequence of  primitives expressed by  a string does not
                           contain enough information to achieve pattern discrimination. For instance, when
                           analysing signals, such as electrocardiograms or heart rate tracings, not all peaks
                           described  by  strings  such  as  x  =  ud  or  x  =  du  are  relevant,  i-e.,  meet  the
                           requirements  necessary  for  attributing  them  to  a  certain  class  of  waves.  Extra
                           conditions often have to be  imposed on the duration and amplitude of  the peaks.
                           We  deal  with  this  requirement through  the  formalism of  attributed strings, i.e.,
                           strings that have an associated attribute vector:




                              For waveforms described by line segments the attribute vector usually has two
                           elements, namely  duration and height of  each segment. In  some cases there  are
                           categorical attributes (eg, colour), and the attribute vector will contain labels (e.g.,
                           red). In  the case of  strings with  single labels, we will  refer to  them as labelled
                           strings.


                            6.2.2  Graphs

                            A graph G is an ordered pair:




                            where N, the node set, is  a set of  nodes, and R,  the edge set, is a set of  binary
                            relations defined in NxN.
                              The elements of R represent arcs (or edges) connecting nodes of N.  An arc of G
                            is denoted (a, b) with a, b E N
                              A  directed  graph,  or  digraph,  is  a  graph  where  each  arc  has  a  direction,
                            emanating from a node and incident onto either another node or the same node. An
                            arc from a digraph, represented by  the pair  (a, b), means that a is the emanating
                            node and b is the incident node. Generally, in a digraph, (a, b) will mean a different
                            relation than (b, a).
                              The number of nodes b, such that (a, b) E  G, i.e.,  the number of emanating arcs
                            from a, is the out-degree of  node a. The number of nodes a, such that (a, b) E G,
                            i.e., the number of arcs incident in b, is the in-degree of node b.
                              Graphs are a  very  flexible way  of  expressing relations among primitives. As
                            there are attributed and labelled strings, we can similarly have attributed or labelled
                            graphs. Figure 6.4 illustrates the use of  digraphs in  character recognition. At  the
                            left  side  of  the  figure,  primitives  and  relations  describing  capital  letters  are
                            indicated. At the right side of the figure, two examples are shown of digraphs using
                            these primitives and relations, for the letters "R" and "EM.
   255   256   257   258   259   260   261   262   263   264   265