MS-TDNN with Global Discriminant Trainings

Émilie Poisson Caillault (1) , Christian Viard-Gaudin (1) , Abdul Rahim Ahmad (1)

1 IRCCyN - Institut de Recherche en Communications et en Cybernétique de Nantes

This article analyses the behavior of various hybrid architectures based on a multi-state neuro-Markovian scheme (MS-TDNN HMM) applied to online handwriting word recognition systems. We have considered different cost functions, including maximal mutual information criteria with discriminant training and maximum likelihood estimation, to train the systems globally at the word level and also we varied the number of states from one up to three to model the basic hidden Markov models at the letter level. We report experimental results for non-constrained, writer independent, word recognition obtained on the IRONOFF database.

Mots clés: Hidden Markov models; Artificial neural networks; Handwriting recognition; Maximum likelihood estimation; Character recognition; Mobile communication; Personal digital assistants; Bayesian methods; Robustness; Power generation

Domaines: Traitement du texte et du document

Émilie Poisson Caillault, Christian Viard-Gaudin, Abdul Rahim Ahmad. MS-TDNN with Global Discriminant Trainings. Eight International Conference on Document Analysis and Recognition, ICDAR, Aug 2005, Seoul, South Korea. pp.856-860. ⟨hal-00337743⟩