Using a large enough dataset is so important in a neural network based handwriting recognition solutions. On the other hand, producing natural datasets is so complicated and time-consuming. In this paper, we compare the effect of using synthesized datasets and natural datasets in state of the art neural network based solutions. For natural solution based datasets we did our experiments by Persian handwriting datasets as Sadri and Yeganeh Zad dataset. The algorithm we have proposed for comparing the effect of datasets in the results is a simple recurrent neural network. The results have shown that using synthesized datasets are as same as natural datasets. In addition, expanding synthesized datasets is so easier and using larger datasets will cause to better accuracy.
Keywords:
Handwriting Recognition, Deep Learning, Natural Dataset, Synthesized Dataset.
