OCR

This project refers to the handwritten recognition with CNN and RNN, decode with CTC.

Demo on youtube

Update

This repo has better model and higher accuracy

Dataset

IAM dataset

• [Model1 - word_model.png] Train on WORD unit of dataset.
• [Model2 - line_model.png] Train on LINE unit of dataset.
• Split data: https://www.openslr.org/56/

Result

Test on IAM dataset:

Model	Test Unit	CER(%)	WER(%)
WORD	WORD	17.20	37.78

Train

[Google colab]

Usage

Training

$ python3 Train.py

Predict

$ python3 Prediction.py

Evaluation

$ python3 EvaluationWord.py
$ python3 EvaluationLine.py

References

https://towardsdatascience.com/build-a-handwritten-text-recognition-system-using-tensorflow-2326a3487cd5 http://norvig.com/spell-correct.html