Contribution to automatic handwritten characters recognition. Application to optical moving characters recognition
Creators
- Gokana, Denis
- Universite de Paris-Sud Centre d'Orsay (France)
- Centre d'Etudes Nucleaires de Saclay, Institut de Recherche Technologique et de Developpement Industriel, Departement d'Electronique et d'Instrumentation Nucleaire - D.E.I.N., Service d'Instrumentation pour la Recherche - S.I.R., Groupe intelligence artificielle et robotique (France)
Description
This paper describes a research work on computer aided vision relating to the design of a vision system which can recognize isolated handwritten characters written on a mobile support. We use a technique which consists in analyzing information contained in the contours of the polygon circumscribed to the character's shape. These contours are segmented and labelled to give a new set of features constituted by: - right and left 'profiles', - topological and algebraic unvarying properties. A new method of character's recognition induced from this representation based on a multilevel hierarchical technique is then described. In the primary level, we use a fuzzy classification with dynamic programming technique using 'profiles'. The other levels adjust the recognition by using topological and algebraic unvarying properties. Several results are presented and an accuracy of 99 pc was reached for handwritten numeral characters, thereby attesting the robustness of our algorithm. (author)
Abstract (French)
Cette these presente un travail de recherche en vision par ordinateur concernant la definition d'un systeme de vision, capable d'identifier des caracteres manuscrits isoles sur un support horizontal mobile. La technique employee consiste en l'analyse de l'information contenue dans les contours du polygone circonscrit au caractere. Apres leur extractions, ces contours sont decrits par des chaines de segments de droite pour donner lieu a une nouvelle methode de representation des caracteres; - la representation a l'aide des profils gauche et droit et des invariants algebriques et topologiques. Nous presentons egalement une methode de reconnaissance par arborescence multiniveaux. Le premier niveau effectue une classification floue munie d'une comparaison dynamique entre les profils par application d'une distance definie par un codage des orientations des contours. Les niveaux suivants raffinent la decision par utilisation des invariants algebriques et topologiques. Des resultats concrets sont presentes et nous atteignons des taux de reconnaissance superieurs a 98 pc dans les cas des caracteres numeriques manuscrits, demontrant ainsi la puissance de notre algorithme. (auteur)Files
45084304.pdf
Files
(45.2 MB)
| Name | Size | Download all |
|---|---|---|
|
md5:93273c562eda8af75655bbcd0c74b6ce
|
45.2 MB | Preview Download |
Additional details
Additional titles
- Original title (French)
- Contribution a la reconnaissance automatique de caracteres manuscrits. Application a la lecture optique de caracteres sur supports mobiles
Identifiers
Publishing Information
- Imprint Pagination
- 196 p.
- Report number
- CEA-N--2502
INIS
- Country of Publication
- France
- Country of Input or Organization
- France
- INIS RN
- 45084304
- Subject category
- S97: MATHEMATICAL METHODS AND COMPUTING;
- Resource subtype / Literary indicator
- Thesis
- Descriptors DEI
- ACCURACY; ALGORITHMS; CLASSIFICATION; COMPUTER CALCULATIONS; COMPUTERIZED SIMULATION; CONFORMAL MAPPING; FUZZY LOGIC; IMAGE PROCESSING; IMAGES; MULTILEVEL ANALYSIS; PATTERN RECOGNITION; PERFORMANCE; SHAPE; TOPOLOGY; VISION
- Descriptors DEC
- MAPPING; MATHEMATICAL LOGIC; MATHEMATICS; PROCESSING; SIMULATION; TOPOLOGICAL MAPPING; TRANSFORMATIONS
Optional Information
- Notes
- 77 refs.; Available from the INIS Liaison Officer for France, see the 'INIS contacts' section of the INIS website for current contact and E-mail addresses: http://www.iaea.org/inis/Contacts/