Two-stage classification system combining model-based and discriminative approaches

For the tasks of classification, two types of patterns can generate problems: ambiguous patterns and outliers. Furthermore, it is possible to separate classification algorithms into two main categories. Discriminative approaches try to find better separation among all classes and minimize the first type of error. But, in general, they cannot deal with outliers. Besides, model-based approaches make outlier detection possible but are not sufficiently discriminative. Thus, we propose to combine a model-based approach with support vector classifiers (SVC) in a two-stage classification system. Another advantage of this combination is reducing the principal burden of SVC: the processing time necessary to make a decision. Finally, experiments on handwriting digit recognition have shown that it is possible to maintain the accuracy of SVCs, while decreasing complexity significantly.