Enhancing Bangla Language Text Recognition in OCR Using Levenshtein Distance and Character Grouping Strategy

One crucial technology for text digitization from image is optical character recognition (OCR). Character recognition in languages with complex scripts, like Bangla, is still difficult despite recent advancements. This research proposes an innovative approach to enhance OCR's Bangla text recognition capabilities. In order to improve recognition performance, OCR errors are efficiently corrected and performance is improved by first applying the Levenshtein distance measure. Experiments show a notable improvement, increasing the initial weighted F1-score of Tesseract OCR from 84.96% to 87.26%. In addition, a method of character grouping is demonstrated to address nuances specific to the Bangla script. Characters that share visual similarities are categorized into different groupings. Subsequently, priority is given to dictionary candidate words with characters that belong to the same class as those identified by OCR. Integrating the character grouping technique yields an astounding 90.92% F1-score.