Word and Diacritic Segmentation Technique Used for Arabic Handwritten Recognition System

Poor recognition rate in most of optical character recognition system OCR is mainly attributed to failing in segmenting a handwritten text correctly, regardless of how well the previous and following stages are designed. We can say that a considerable share of recognition errors is attributed to the segmentation stage. In this paper a new algorithm for segmenting handwritten Arabic text-line into words and diacritics is proposed. In our approach we rely on using fill-in function, stroke thickness, text-line height in segmenting the words/subwords and diacritics of Arabic handwriting. We have developed a simple procedure for stroke thickness estimation. In addition to segmentation, the proposed algorithm performs partial recognition as it can recognize the Arabic character Alif and some diacritics during the segmentation process. Higher segmentation accuracy was obtained from number of conducted experiments