Beatrix: A self-learning system for off-line recognition of handwritten texts

Abstract We present a self-learning system, BEATRIX, for off-line recognition of handwritten texts. The system integrates neural recognition with context analysis techniques. It consists of three main interacting subsystems: the first is based on an ensemble of neural networks and carries out an approximate pre-recognition of characters; the second carries out a lexical and grammatical analysis of the recognised text. This analysis produces hypotheses about words and sentences in order to correct errors made by the neural networks. Once a sufficient number of words have been recognized, the third subsystem retrains one of the neural networks with the hypotheses produced. This enhances the capacity of the system to recognise the specific handwriting, without losing the capability to recognise other types of handwriting.