Synthetic Handwritten Gesture Generation Using Sigma-Lognormal Model for Evolving Handwriting Classifiers

International audienceWe show in this paper the importance of using handwriting generation in the context of online and incremental learning of a handwriting classifier. In order to obtain realistic synthetic gestures, we apply controlled deformations on the extracted sigma-lognormal parameters of the real gesture, and we then generate synthetic gestures using the modified parameters. Results show the impact of integrating these synthetic samples generation in our learning algorithm on the classification performance