A new Efficient Graphemes Segmentation Technique for Offline Arabic Handwriting

In this paper, the challenging problem of the segmentation of Arabic handwriting is addressed. We propose a new efficient technique for segmentation of offline Arabic handwriting into the basic graphemes of the handwritten Arabic script. The proposed technique applies the Douglas-Peucker algorithm on the skeletonized parts of the offline handwritten images to convert them into a set of piecewise linear curves where the local writing direction information and the neighborhood geometric characteristics are used for graphemes segmentation. As a pre-processing stage to the proposed segmentation technique, a reliable rule-based method is used for diacritics extraction that is based on the geometric properties and writing positions of those diacritics. The segmentation technique has been tested using 1400 Arabic handwritten words from the IFN/ENIT database and a comparison made against the recent existing methods in the literature and a promising segmentation performance has been achieved, indicating the effectiveness of the proposed technique.