Résumé: The main theme of this work is the segmentation of arabic handwritten words using a multi-agents approach. The proposed system deals with arabic handwriting, namely naskh or its modified version, written by different writers and segments the words into characters. A multi-agent approach is used to define regions where segmentation is illegal. Seven agents are used : six are major and one is secondary. First, the image of the arabic word is preprocessed and thinned before feature points are extracted. Next, the segmentation process is applied by invoking the agents which cooperate with each other to determine the regions where segmentation is not allowed and return the rest of the word for more investigation. A heuristic algorithm is applied then to find the initial cutting points which are filtered based on specific rules to return the final condidate cutting points. Finally, the resulting segments/characters are verified and the candidate segmentation points are validated using a neural network and a set of combination rules to handle the over-segmented characters. The experimental results are very promising for both systems : segmentation and recognition.

Langue: Anglais
Collation: 109 p. ;30 cm.
Diplôme: Master
Etablissement de soutenance: Sharjah, University of Sharjah
Spécialité: Informatique
Index décimal 621 .Physique appliquée (électrotechnique, génie civil, génie mécanique, ingénierie appliquée, principes physiques en ingénierie)
Thème Informatique

Mots clés:
Systeme de reconnaissance (informatique)
Segmentation (Informatique)
Reconnaissance, Méthodes de (Informatique)

Note: Bibliogr. pp.100-109