Synthizing Handwritten Arabic Text Using Active Shape Models

This research paper proposes a template based approach for generating an unlimited number of synthesized handwritten Arabic characters. It starts by generating a polygonal representation for each training sample for every character class. Then for each class an Active Shape Model (ASM) is used to unify all polygonal samples in one compact representation. Accordingly, any desired number of synthesized characters, can be produced, as a result of simple linear combination between the Eigenvalues and Eigenvectors of the ASM. Ultimately, the contour of synthesized character is smoothed using piecewise cubic hermit interpolation. Moreover, by combining multiple synthesized characters, our system is capable of producing synthesized Arabic words. Even though experiments have shown that a perfect human- like handwriting is still far away. We think that our approach is a very promising and a step forward towards achieving this goal.