On-line cursive handwriting analysis and recognition: paradigm and systems

There are two fundamental characteristics in handwriting: the external shape or significant, which characterizes the author of the handwritten text; and the semantic aspect or signification, which indicates the meaning of what has been written. We focus only on handwriting recognition. We have proposed a hierarchical modelling of handwriting lines, words, allographs and ligatures, in terms of visual and dynamic clues. A logical segmentation of word is also considered. A new approach of the recognition, based on robust features only, is described. It consists of three steps: elimination, classification and discrimination. Three systems for cursive letter and word recognition have been overviewed. Two approaches, structural and stochastic, have been considered. The first results are globally encouraging