On the problem of classifying Vietnamese online handwritten characters

Recognizing circumflexes and diacritics in a Latin-based tonal language of Vietnamese is important because they are used to distinguish words. There are two main approaches to solve the diacritics recognition problems. The first approach tries to separate diacritics from their main character and recognizes them independently. This approach, however, should only be applied with printed documents as the diacritics of printed characters are separable. With free-styled handwritten characters, diacritics are very hard to separate and recognize because they may overlap main characters. The second approach tries to recognize both main character and diacritics without separation step. In this paper, we want to show that our approach with an appropriate feature extraction method and subclass diving strategy, a robust recognition can be obtained with high recognition rate.