The recognition graph - language independent adaptable on-line cursive script recognition

One of the difficulties involved in providing commercially acceptable cursive script recognition solutions has been that the training of most systems have been greatly dependent upon dictionaries. In this paper we present a new technique for dictionary interaction in on-line cursive script recognition. It retrieves all possible paths through a segmentation graph, that correspond to words in a dictionary, in a very efficient way. Furthermore we present new ideas of how to treat secondary strokes in the on-line segmentation graph. The novel system has been tested on large data with very competitive results.