Transfer Learning to improve Arabic handwriting text Recognition

In recent years, the leveraging of deep learning approaches allows a great progress in text recognition task. But they usually need a considerable amount of training examples to learn a new model. Therefore, lack of data can be an issue when developing a new recognition model, especially for handwriting Arabic text recognition where the lack of databases is stilling an interested problem. In this context, the main contributions of this paper is based on transfer learning the parameters learned with a bigger mixed-fonts printed Arabic text database to handwriting one. Experiments shows the good improvement provide with this technique.