Abstract
In this paper an improved method of printed Arabic character recognition is presented. It is segmentation-free character recognition. A sliding widow with the size of a reference character is used to select a sub-image, from the document image, for the recognition. The proposed procedure starts with highest probability characters in Arabic writing, except for those characters that produce a recognition error. To make the procedure even faster, characters that have same dimensions are grouped in one cluster and recognized together. A previously created database for Arabic characters images is used in this research as reference characters. Five font types and nine font sizes are implemented. The classifications of the unknown characters are carried out by extracting their structural and transform features. The Character height, width, and number of pixels above baseline, together with the Walsh Hadamard Transform (WHT) coefficients are used to construct the features vector.