Handwritten Text Recogniton
Rather than segmenting a word into characters and recognizing the characters, we have adopted the approach of looking for every character in a word and based on detections identifying the resulting string.
This works well for overlapping characters, and does not require a one-dimensional arrangement of the characters. This was developed and tested initially for the Bangla script, but has also been successfully tested on English and Korean. A dataset of Bangla text has been released. Currently open source code and English YOLO weights are being developed. Other enhancements in progress include adding techniques to recognize and assign diacritics and to evaluate its performance on crossed out text.
Publications:
Nishatul Majid, Elisa H Barney Smith, “Character spotting and autonomous tagging: offline handwriting recognition for Bangla, Korean and other alphabetic scripts,” International Journal on Document Analysis and Recognition (IJDAR), pp 1-19, 2022.
R Mondal, S Malakar, Elisa H Barney Smith, R Sarkar, “Handwritten English word recognition using a deep learning based object detection architecture,” Multimedia Tools and Applications, v 81, n 1, p 975-1000, January 2022
N Majid, EH Barney Smith, "Segmentation-free Bangla offline handwriting recognition using sequential detection of characters and diacritics with a Faster R-CNN," International Conference on Document Analysis and Recognition (ICDAR), 2019
N Majid, EH Barney Smith, "Introducing the Boise State Bangla Handwriting dataset and an efficient offline recognizer of isolated Bangla characters," 16th International Conference on Frontiers in Handwriting Recognition (ICFHR), Niagara Falls, New York, USA, 2018
Participants:
Nishatul Majid (Fort Lewis College), Elisa Barney Smith (LTU)
Updated: