| Author(s) | Asim Abdul-Fattah Nabawi *; Sabri Abdullah Mahmoud ** |
| Affiliation | * Computer Engineering Dept. cC/So King Saud Unil'ersity. Riyadh ** Al-Manarain Est. for Technical Applications. P.G. Box 53531. Riyadh 11593 |
| Title | Recognition of Typewritten Arabic Characters Using |
| Source | Journal of King Saud University. Computer & Information Sciences. Volume 9, No 1. (1997/1417) |
| Abstract | Research efforts in the field of Arabic character recognition using optical techniques have intensified in the recent past. This state of affairs prevail due to several reasons, among which is the increasing interest in computer systems combined with their affordable availability to a broader class of users, not to mention the technical advances in OCR applications in languages other than Arabic. This paper addresses the optical recognition of Arabic characters using the well known backpropagation neural network. Features of the Arabic characters based on Walsh transform and the moments method were extracted. Both classical and neural network classifiers were used in both the training and recognition phases. Comparison of the two classification techniques and the two types of extracted features was carried out. The results of this research work have shown that the use of the Walsh-transform- based features yielded higher recognition rates in addition to faster processing times compared to those of the moments-based technique. The latter technique, however, requires less memory. The backpropagation technique was found to be more tolerant to noise than the classical classifiers. The former, however, was found to require several trials to adapt its architecture to the needs of the particular application. The experimental results have ascertained the effect of using larger sets of characters on improving the recognition rates in all instances. |
