There has been good amount of discussion about OCR lately. Since existing open-source OCR solutions doesn’t provide the level of accuracy and training feasibility like their commercial counterparts we have started to look towards Deep Learning based solutions and we have found some interesting works.
Calamari − A High-Performance Tensorflow-based Deep Learning Package for Optical Character Recognition
The authors have provided code to replicate the results in the paper, they aim to provide a more robust open-source OCR solution. It is designed to both be easy to use from the command line but also be modular to be integrated and customized from other python scripts.
Calamari OCR – GitHub repo
Calamari Paper – Paper
Calamari in its present form expects input as an image of text line, so the input page has to be segmented into individual lines before being fed into calamari.
Text Line segmentation using Deep Learning
Couple of deep learning based solutions using fully convolutional neural networks have been proposed. The results provided in the papers are promising.
Calamari combined with text line segmentation has the potential to be more robust than existing open-source OCR solutions.