A Progressive Learning Method for Symbol Recognition

This paper deals with a progressive learning method for symbol recognition which improves its own recognition rate when new symbols are recognized in graphic documents. We propose a discriminant analysis method which provides allocation rules from a training set of labelled data. However a discriminant analysis method is efficient only if the training set and the test data are defined in the same conditions but it is rare in real life. In order to overcome this problem, a conditional vector is added to each instance to take into account the parasitic effects between the test data and the training set. We also propose an adaptation to consider the user feedback.