Spectral Graph Matching for Printed Tifinagh Character

Optical Character Recognition is one of the most important tools that contributes to facilitate man-machine interaction. In this paper, we present an optical Tifinagh character recognition system based on graph theory. After preprocessing, interest points are extracted using Harris corner detector. Based on these points, we constructed the graph model representation of Tifinagh characters. Classification is done by calculating the spectral properties of adjacency matrix that represent the degree of agreement between graphs. The system shows satisfying performance and robustness against noise and deformation. The proposed system is evaluated using IRCAM database (Royal Institute of Amazigh Culture) and recognition rate of 99.02% was obtained.