Training and evaluation of the models for isolated character recognition

Overview & Synopsis

In this report we describe a special multi-layer perceptron wich was trained on the image database in order to obtain the classification of the test database. We used the same database to train for each class an SVM and after that we used a normalization process in order to be able to compare the different results given by the SVMs on the test database, as they are not n-class classifiers. As the results obtained were not enough satisfactory for our industrial partner, we proposed a combination scheme, in order to interact the different decisions given by the different classifiers and to obtain a high recognition score.