New Morphological Markovian Approach for Analysis and Recognition of Open Arabic Canonical Vocabulary

摘要

Arabic writing recognition has been a tough challenge due to its flexional nature and great topological variability. For that, we have been continuing investigating the use of linguistic knowledge to improve the recognition of wide then open Arabic word lexicon. In this paper, we propose a new approach for the recognition of decomposable Arabic words by the use of planar (i.e. bi-dimensional) Markovian models that embody two crucial aspects of Arabic: language morphology and script topology. The contribution here aims to extend the target from wide lexicon recognition toward open lexicon one. Indeed, for the training, we use planar hidden Markov models; each is dedicated to learn a sub-vocabulary derived from one root. For the recognition, we use already trained models, for non-learned words, we instantly created a new model which is able to recognize words derived from new intruder (non-trained) roots. Preliminary experiments were conducted on a corpus of about 3000 samples of Arabic words and yielded promising results.