Abstract
This paper proposes a new structural approach of features extraction for isolated Arabic characters based on the "Hough Transform". The new approach consists first in detecting loops and points in the structure of the entered Arabic character, and classifying it accordingly. Then the "Hough Transform" is used to detect the longest lines in the character structure, in different directions, and according to a well-chosen threshold. Afterwards, we have made improvements to this method, by introducing another threshold to consider also the shortest lines in the character structure. A recognition rate of 99% has been achieved for the recognition of Arabic characters written with fonts that have already been learned in the learning phase, and a remarkable rate for characters written with other fonts than those used in the learning phase.