R.J. Qureshi, J.-Y. Ramel, and H. Cardot (France)
Document image analysis, graphic symbols recognition, attributed relational graph matching.
Graph representation and graph matching have been successfully applied to a large number of problems in computer vision and pattern recognition. Concerning graph matching, the classical algorithms of graph isomorphism seems useless when the image is degraded with noise or vectorial distortion. This paper introduce a novel similarity measure to recognize symbols by performing inexact matching of attributed graphs. In the proposed approach, symbols are encoded by attributed graphs, whose nodes represent structural primitives like quadrilaterals and whose edges represent mutual relationships between these primitives. To be invariant of rotation and scaling, relative information about primitives are associated as attributes on the nodes and edges. Considering a mapping between two graphs, a similarity function is formulated, that use the numerical values of the attributes to calculate a similarity score. This new similarity measure has many desirable properties such as discrimination power, invariant to affine transformations, and robustness to noise or vectorial distortions.
Important Links:
Go Back
