In this paper a novel feature extraction methodology for character recognition by means of the information in inverse order stroke information using wavelets is proposed. The information used to reconstruct and recognize the stroke is done by using the so-called optimal knots. In this paper 20 optimal knots were used. As approximating function we decided to use the slalom natural spline function. The recognition experiments were carried out using the obtained feature vector as the input to some recognition systems based on Neural Network, Support Vector Machines and Gaussian Mixture Models for comparisons purposes. The proposal was evaluated with a database of seven writers with 50 traces of each English character. The global recognition rate when using the three recognition strategies varied between 98 and 98.7 %.
|Original language||American English|
|Number of pages||154|
|Journal||Revista Facultad de Ingenieria|
|State||Published - 1 Sep 2009|