Neuro-classification for handwritten Arabic text. (c2002)

Abstract

The problem of handwritten character recognition is still a big challenge for the scientific community. Several approaches have been attempted with that purpose in the last years, but most of these focused on the English pre-printed or handwritten characters space. For this purpose, this research has been done to shed some light over the Arabic handwritten text recognition. Yet, several major problems attach to Arabic handwritten text processing: "Arabic is written cursively and many external objects are used such as dots, ' Hamza', 'Madda', and diacritic objects; Arabic characters have more than one shape according to their position inside a word". These problems result in a difficult classification solution. Due to the fact that algorithms based on neural networks have been proved to give better results than conventional methods, when applied to problems where the decision rules of the classification problem are not clearly defined, two neural networks were built to classify already segmented characters of handwritten Arabic text. The process starts by using a neuro-conventional algorithm, and a neural network for segmenting binarized connected blocks of characters (Hamid 2001] [Zabadani 2002], and then a heuristic algorithm extracts features from these characters and feed them into two neural networks for classification purpose. One hurdle was encountered in the above scenario, which can be summarized as follows: there are a lot of handwritten characters that can be segmented and classified into two or more different classes depending on whether you look at them separately, or in a word, or even in a sentence. In other words, character classification, especially handwritten Arabic characters, depends largely on contextual information, and not only on topographic features extracted from these characters.