5 - Utilisation des processus markoviens en reconnaissance de l'écriture

In this paper, we present a brief survey on the use of different types of Markov models in writing recognition . Recognition is done by a posteriori pattern class probability calculus . This computation implies several terms which, according to the dependency hypotheses akin to the considered application, can be decomposed in elementary conditional probabilities . Under the assumption that the pattern may be modeled as a uni- or two-dimensional stochastic process (random field) presenting Markovian properties, local maximisations of these probabilities result in maximum pattern likelihood . We have studied throughout the article several cases of subpattern probability conditioning. Each case is accompanied by practical illustrations related to the field of writing recognition .