Identity Signatures Extraction of Latin and Arabic Characters

Abstract

We present an original method of identity signatures extraction of online handwritten Arabic and Latin characters based on the frequent pattern mining, through the “Apriori” algorithm. This work is based on models that we have generated, to describe the writer behavior that illustrates a trajectory, from local and global features which include temporal information. We also manipulated the minimum threshold value, which is an essential parameter in frequent patterns extraction algorithms, to obtain the dominant spatiotemporal features of the character identity. In addition, we have incorporated the patterns frequency called “support” as a new statistical feature relevant in the signature. These signatures can also provide portability and add a great flexibility to systems that adopt this type of modeling. The obtained results are very promising as we have achieved very high rates of correct recognition and the models which generated these results are considered as relevant and reliable models.