Year: 2024
Type: conference-paper
Abstract: The significance of automatic Bengali Handwritten Text Recognition (HTR) is increasingly recognized, given its pivotal role in diverse domains like document digitization, intelligent traffic systems, and postal automation.We've created a novel dataset meticulously curated to the highest standards, entitled BN-HW-DSNd.This dataset boasts comprehensive labelling and a vast reservoir of data, offering a diverse range of modalities.Despite the inherent challenges posed by the variations in size and shape of handwritten characters, particularly in the curvy and compound Bengali script, our study also introduces a Convolutional Neural Network (CNN) based image classification model tailored for recognizing Bengali handwritten district names.With a collection of 7,040 handwritten samples of various shapes and colours, our designed model showcases outstanding performance.It achieves a test accuracy of 98.44%, showcasing the efficacy and sensitivity of our model in addressing this intricate task.
Institution: Premier University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4411099302