Developing a Bangla Handwritten Text Recognition Framework using Deep Learning

There are 250 million speakers of Bangla, the second most common language on the Indian subcontinent, making it one of the most significant tongues in the world. In this highly populated subcontinent of over 300 million people, Bangla is the sixth most utilized language in the world. Despite its popularity, the slow progress in identifying handwritten Bengali writing is depressing. Bangla handwritten script’s complexities, which include compound characters, intricate punctuation, and cursive writing styles, provide difficult hurdles. Recent breakthroughs in machine learning, particularly Convolutional Neural Network (CNN)-based architectures, have shown amazing accuracy in Bangla Handwritten Text Recognition. This study aims to provide a fresh and distinct Bangla handwritten text recognition model by using the capabilities of the "Bangla Handwriting Dataset." This extensive dataset includes 260 single-page handwritten samples from people of all ages and personalities. The word labels and boundary boxes in the dataset were meticulously created. This research emphasizes using cutting-edge tools and technologies, uniting them under a unified framework. Among the different CNN-based architectures investigated, DenseNet121 outperforms competitors such as Xception, MobileNet, MobileNetV2, and NASNetMobile for text recognition. Furthermore, when combined with a GRU RNN layer, DenseNet121 obtains the most remarkable test accuracy of 95.643%. A prototype with a user interface has been created to improve accessibility and user-friendliness. This program quickly scans photos of manuscripts, interprets each word, and turns them into editable text. This comprehensive technique for recognizing Bangla handwritten text promises to open new channels of communication and information access in a linguistically varied and technologically linked society.