Contribution to the recognition of hand Arabic word based on neural network

This work describes an automatic recognition system for the handwritten Arabian words (as application we takes the 48 cities). The major problem in the automatic recognition of the cursive Arabian writing is the segmentation in their different constituent. First of all, we classify the cities along the number of the related components, then we segment every component in characters. For the segmentation in characters we proposed a method that essentially rests on the skeletons of the pictures of the cities and the detection of the points of branching or crossing (calls them the essential points). After the segmentation, we tried to describe the characters by their applicable features that permit to classify them to best, for it we chose some topological features (detection of the holes, the concavities in the four directions, the numbers of the diacritical points and their positions, ...)