1998 Fiscal Year Final Research Report Summary
PEN-BASED HANDWRITING RECOGNITION SYSTEM OF THE ARABIC SCRIPT
| Project/Area Number |
09041129
|
|---|
| Research Category |
Grant-in-Aid for international Scientific Research
|
| Allocation Type | Single-year Grants |
|---|
| Section | Field Research |
|---|
| Research Field |
Intelligent informatics
|
|---|
| Research Institution | Hiroshima City University |
|---|
Principal Investigator |
BOUSLAMA Faouzi Hiroshima City University, Department of Computer Science, Associate Professor, 情報科学部, 助教授 (20261240)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
AMIN Adnan The University of New South Wales, School of Computer Science and Engineering, A, School, 助教授
BEBREJEB Mohamed The University of Tunis II,Ecole Nationale d'Ingenieurs de Tunis-ENIT,Tunisia, P, ENIT, 教授
SANO Manabu Hiroshima City University, Department of Computer Science, Professor, 情報科学部, 教授 (10092785)
|
|---|
| Project Period (FY) |
1997 – 1998
|
| Keywords | Character Recognition / Arabic Characters / On line / Structural Approach / Feature Extraction / Classification / Fuzzy Logic / NeuralNetworks / Off line |
|---|
| Research Abstract |
This research project is on the automatic recognition of handwritten and machine printed online and offline Arabic characters. New hybrid approaches based on classification techniques such as structural and statistical approaches combined with soft computing techniques of fuzzy logic and neural network methods were proposed as new methods of classification. For online systems, Arabic characters were drawn on a grahic tablet and simple features corresponding to the geometric characteristics of characters were chosen. With the use of fuzzy logic based modeling techniques, the complex handwriting styles and variations were described in a broad way, thus reducing the difficulties in system formulation and generating a flexible and general solution to the recognition problem. In offline systems, various techniques were proposed and discussed. Structural features and fuzzy classifying rules gave very high recognition results. Statistical features and neural networks were also used and gave satisfactory results.
|
