Segmentation of Arabic characters using their contour information

Arabic characters are written cursively in both printed and handwritten forms and they consist of 28 main characters. However, most of their shapes change when they are joined to form a word. This poses a challenging problem if these characters are to be recognised. We present an algorithm which allows Arabic words to be segmented into the proper set of characters. This is achieved by choosing the right position of segmentation which is indicated by an angle formed between each pair of joined characters. This algorithm is superior to those that rely on the histogram method. It improved the recognition rate of the segmented characters to 97.01% over a set characters consisting of two fonts each in four different sizes.