Handwritten Text Recognition: A Deep Learnin...

Handwritten Text Recognition: A Deep Learning Approach with CNN-BiLSTM Architecture

Author Name : Anusha R, Disha M Hulkund, Gayathri K R, Harshitha Atrey, Dr. Shantha Kumar H C

DOI: https://doi.org/10.56025/IJARESM.2025.1301253176

ABSTRACT Handwritten text Recognition (HTR) is a challenging task, as the idiosyncratic nature of writing provides great variability between different people, as well as different types of fonts, noise injected into images, and distortions added by writing tools or by the scanning process. In this work, we focus on identifying a stronger HTR system built on Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks. The model is intended to understand different styles of writing and to convert the structured images of the input text to readable machine text. The system streamlines the digitization process for handwritten documents, unlocking new use cases in healthcare, education, and government, by addressing these challenges. On the IAM Handwriting Dataset, it shows remarkable accuracy improvements and holds prospects of realworld applications. Future work also involves broadening the support for multi-lingual text recognition & optimizing.