A Transfer Learning Approach for Handwritten Numeral Digit Recognition

Handwritten numeral digit recognition is a classical problem in the field of computer vision, which has a wide range of applications in various fields including financial and post services. The accuracy of handwritten numeral digit recognition has been greatly improved by using deep learning in the past few years. However, deep learning relies on a large amount of training data and time-consuming calculation. In this paper, we adopt a transfer learning approach for handwritten numeral digit recognition and use both the multi-layer perceptron and convolutional neural network models to share the feature extraction process among five handwritten numerical datasets, namely, Tibetan, Arabic, Bangla, Devanagari, and Telugu. We compare the transfer learning scheme with the model based on a single dataset. We find that using the transfer learning method can significantly reduce the training time of the deep learning models, and slightly reduces the recognition accuracy.