Deep Learning Based Text Extraction System from Handwritten Bengali Word

Extracting text from handwritten Bengali images poses unique challenges for Optical Character Recognition (OCR) systems due to varied handwriting styles, diverse character representations, alignment issues, character similarities, and distinguishing marks. By overcoming these challenges, our research contributes to advancing OCR technology and facilitates improved text extraction from handwritten Bengali documents. We developed an intelligent system that uses Deep Learning to localize and classify Bengali characters. Post-processing techniques are then applied to convert these characters into readable text. Our system encompasses 82 classes, including vowels, consonants, diacritics, numbers, and compound characters. Evaluation shows promising results, with an average accuracy of 91% for character localization and classification, and an overall system accuracy of 86.59%.