An application of Vietnamese handwriting text recognition for information extraction from high school admission form

Abstract

This paper presents an effective Vietnamese handwritten text recognition model by applying an improved convolutional recurrent neural networks (CRNNs) model to high school enrollment forms in Tay Ninh province, Vietnam. First, the proposed model extracts data areas containing text characters from forms. Then, we connect text boxes on the same row and divide the fields that containing text into three specific regions. Finally, we detect areas containing text characters for handwritten text recognition. We use word error rate (WER) to evaluate the recognition process and obtain a result of 0.3602. This result is one of the best solutions to the Vietnamese handwritten text recognition problem.