Segmentation of Handwritten Arabic Words Using High Level Informative Scheme

In handwritten Arabic word recognition, the segmentation is a vital and crucial phase, and has a great impact in the accuracy of the subsequent recognition phase. We propose in this paper a new segmentation algorithm based on a high level informative scheme, which describes and identifies different geometric forms in a word, and use them to segment words in characters. The informative scheme we propose gives us the description of skeleton points, their coordinates, natures and number of their neighbors in 8-neighborhood. This allows the extraction of representative primitives of characteristics points that are used in the segmentation phase. The experimental results are promising, and can be improved; we obtained an important number of Good Segmentation Points and a low number of Bad Segmentation Points.