Hybrid CNN-GRU Model for Handwritten Text Recognition on IAM, Washington and Parzival Datasets

Abstract

The aim of using the CNN-GRU Hybrid Model for HTR is to improve the accuracy of existing HTR systems by developing more robust models that can handle the variability of handwriting styles and the complexity of language. The proposed model combines CNN and GRU and is evaluated on multiple datasets, including IAM, Washington, and Parzival, to provide a comprehensive analysis and comparison with existing models. The CNN-GRU architecture proposed in the study has been tested on IAM, Washington, and Parzival datasets, and it was found to have lower CER and WER scores compared to many other models. The model achieved CER scores of 7.16%, 6.S%, and S.06% and WER scores of16.16%, 17.24%, and 19.13% on the IAM, Washington, and Parzival datasets, respectively.