Skip to Main content Skip to Navigation
Theses

Modèles markoviens uni- et bidimensionnels pour la reconnaissance de l'écriture manuscrite hors-ligne

Abstract : This thesis deals with the study of stochastic Markovian modelings of variable nature and dimension and with their application to off-line handwriting recognition within a small lexicon. For the ID models, a comparison is performed between the HMMs (discrete and semi-continuous) and the stochastic trajectory model (STM). The pseudo-2D or planar-HMMs are an important step towards the extension of the warping to the plane. Next, the accent is set on the use of causal random fields where we introduce a new model called NSHP-HMM. The properties of this model and the experiments performed on the recognition of characters (printed and handwritten), handwritten digits and unconstrained handwritten words in a small lexicon are described in great detail. The study of training issues and recognition methods is done for all the modelings considered. Some elements related to classifier combination and reject are also described. Complexity considerations together with practical implementation issues show the feasability of the proposed models within a real application.
Document type :
Theses
File URL :
http://docnum.univ-lorraine.fr/prive/SCD_T_1997_0299_SAON.pdf
Complete list of metadata

https://hal.univ-lorraine.fr/tel-01747325
Contributor : Thèses Ul <>
Submitted on : Thursday, March 29, 2018 - 10:57:20 AM
Last modification on : Friday, February 26, 2021 - 3:26:01 PM

Identifiers

  • HAL Id : tel-01747325, version 1

Collections

Citation

George Saon. Modèles markoviens uni- et bidimensionnels pour la reconnaissance de l'écriture manuscrite hors-ligne. Autre [cs.OH]. Université Henri Poincaré - Nancy 1, 1997. Français. ⟨NNT : 1997NAN10299⟩. ⟨tel-01747325⟩

Share