Yasin Zamani - Department of Computer Engineering, Sharif University of Technology, Tehran, Iran
Yaser Souri - Department of Computer Engineering, Sharif University of Technology, Tehran, Iran
Hossein Rashidi - Department of Computer Engineering, Sharif University of Technology, Tehran, Iran
Shohreh Kasaei - Department of Computer Engineering, Sharif University of Technology, Tehran, Iran

Persian handwritten digit recognition has attracted some interests in the research community by introduction of large Hoda dataset. In this paper, the well-known random forest (RF) and convolutional neural network (CNN) algorithms are investigated for Persian handwritten digit recognition on the Hodadataset. Using the Hoda dataset as a standard testbed, we have performed some xperiments with different preprocessing steps, feature types, and baselines. It is then shown that RFs and CNNs perform competitively with the state-of-the-art methods on this dataset, while CNNs being the fastest if appropriate hardware is available.

Machine learning, Random forest, Convolutional neural network, Handwritten digit recognition, Persian digits, Hoda dataset