Abstract:
This thesis presents a novel system approach of online Arabic
handwriting recognition as well as datasets of online Arabic handwriting.
To fill the gap and shortage of standard online Arabic handwriting
datasets, the thesis introduces two datasets of this handwriting. The XML
representation is selected for the datasets design. The construction of the
datasets is achieved by developing two software tools (collection tool,
verification tool) and then using these tools to collect and maintain online
Arabic handwriting for the datasets.Therefore, we have acquired standard
datasets which can be used by researchers to train and test their recognition
techniques for this handwriting.
These acquired datasets are used in the training and testing of the system
approach which developed in this thesis. This system involves a position
invariant feature extraction method. Moreover, it involves an algorithm which
specifically designed in this thesis to divide the given cursive words of the
handwriting into segments of Arabic letters. To fulfill the recognition, the
system employs hidden Markov models so as to model the handwriting.The
experiments which performed to evaluate the system have shown that the
system performance is influenced by the feature arrangements within those
feature vectors of the letter segments; such that the well done choice of the
arrangement has enhanced the performance.
