A Study on a Hybrid CNN-RNN Model for Handwritten Recognition Based on Deep Learning

Jun Ma

2023

Abstract

In today’s era of digitalization, the efficient conversion of handwritten content into digital formats remains essential despite the widespread adoption of digital document storage. This study addresses the pressing need for efficient conversion of handwritten content into digital formats. Furthermore, the preprocessing procedures employed on handwritten images, including deskewing and normalization, were delineated. This study embraces a hybrid model-oriented recognition approach by utilizing the proposed hybrid Convolutional Neural Network (CNN)-Recurrent Neural Network (RNN) model for handwritten text recognition. It juxtaposes it with a solitary CNN model. The hybrid model’s central components include a CNN for feature extraction and a Bidirectional Long Short-Term Memory network for sequence modeling. These components work together to enhance the precision of recognizing handwriting text. The research employs visualization techniques to understand the model’s operations and improve performance. The CNN-RNN hybrid model significantly outperforms the CNN model, achieving a 12.04% reduction in Word Error Rate (WER) and a 5.13% Character Error Rate (CER). Conclusions drawn from the study illustrate that the suggested hybrid deep neural network model outperforms the conventional CNN method in terms of handwriting recognition accuracy. This is conducive to advancing the practical application of handwritten text scanning and recognition.

Paper Citation

in Harvard Style

Ma J. (2023). A Study on a Hybrid CNN-RNN Model for Handwritten Recognition Based on Deep Learning. In Proceedings of the 1st International Conference on Data Analysis and Machine Learning - Volume 1: DAML; ISBN 978-989-758-705-4, SciTePress, pages 268-273. DOI: 10.5220/0012801000003885

in Bibtex Style

@conference{daml23,
author={Jun Ma},
title={A Study on a Hybrid CNN-RNN Model for Handwritten Recognition Based on Deep Learning},
booktitle={Proceedings of the 1st International Conference on Data Analysis and Machine Learning - Volume 1: DAML},
year={2023},
pages={268-273},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012801000003885},
isbn={978-989-758-705-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Data Analysis and Machine Learning - Volume 1: DAML
TI - A Study on a Hybrid CNN-RNN Model for Handwritten Recognition Based on Deep Learning
SN - 978-989-758-705-4
AU - Ma J.
PY - 2023
SP - 268
EP - 273
DO - 10.5220/0012801000003885
PB - SciTePress