Graph based recognition of isolated graphic symbols

Graph matching methods are often proposed to solve the symbols recognition problems. Finding an exact matching between two graphs using graph isomorphism techniques remains unresolved and known to be NP-complete. In our proposed technique, first, raster image of each graphic symbol is vectorized and quadrilaterals are constructed by searching and selecting vectors having opposite direction and minimum distance between their extremities. We use quadrilaterals and their neighbouring relations to construct a labelled graph. To deal with the presence of noise and distortion, our approach is based on inexact graph matching where each vertex of one graph can be associated to zero, one or more vertices of the other graph. It uses a greedy incomplete search algorithm to find the best mapping between the vertices of the two graphs and compute a similarity score. The strength of this approach is maximum structural representation which is invariant to distortion and affine transformations.