OFFLINE ARABIC HANDWRITTEN ISOLATED CHARACTER RECOGNITION SYSTEM USING SUPPORT VECTOR MACHINE AND NEURAL NETWORK

摘要

The Arabic Language had a little attention in this field compared with other languages due to the high cursive nature of the handwritten Arabic language, especially with their dots. The difficulty lies in the complexity of locating the wavy shape in the characters, which solved by the combination of certain features extraction methods that work in separate way. The proposed of Isolated Arabic off-line handwritten recognition system based on two stages classifiers (Hybrid). First stage is a linear Support Vector Machine (SVM) for splitting the dataset characters into two groups - Characters with dots and Characters without dots, by giving certain extraction features to each group. This division can reduce the error rate of characters recognition which has similar looking shape. Second stage supplies the first stage result to Neural Network (NN) stage which granted one of the best correctness and accuracy by training. Finally, a fully recognized character is acquired successfully. This work is implemented (IFN/ENIT) dataset, the system significantly reduce the load of NN process by SVM classifier, which can be used for real-time applications. A total accuracy of this proposed work reaches 92.2%.