Abstract:
In this work we present a system for the recognition of handwritten Arabic text using neural networks. This work builds upon previous work done by [Hamid 2001]. That part dealt with the vertical segmentation of the written text. However, faced with some problems like overlapping characters that share the same vertical space, we tried to fix that problem by performing horizontal segmentation. In this research we will use two basic neural networks to perform the task; the first one to identify blocks that need to be horizontally segmented, and the second one to perform the horizontal segmentation. Both networks use a set of features that are extracted using a heuristic program. The system was tested with over 1500 characters (each character has on average about 50 rows) and the rate of recognition obtained was over 90%. This strongly supports the usefulness of proposed measures for handwritten Arabic text.
