Segmentation of handwritten Arabic words

Abstract EN

In this paper, we develop an efficient technique for the segmentation of on-line handwritten Arabic words.

The segmentation process is followed by the recognition of the segmented characters and consequently recognizing the word.

Some words contain one subword while others may contain two or more subwords.

Thus each word, which may contain one character, is independently processed.

The technique detects whether the subword contains one character or more.

For the first case, no segmentation is required, while segmentation is necessary in the other case.

Generally, a subword contains one main stroke and one or more secondary strokes.

For the segmentation ofthemain stroke of a subword, special cases for the first character is considered first.

Consequently, the last character of the subword is checked for special cases also.

Having considered all special cases, other general cases of characters that are within the subword or at its beginning or end are considered in a different manner.

The segmentation technique has been applied to a large set containing about one thousand subwords with a probability of success of about 0.99.

This high probability is mainly due to the fact that the tested words have been entered by a single writer and were not completely free.