OCR
This project refers to the handwritten recognition with CNN and RNN, decode with CTC.
Update
This repo has better model and higher accuracy
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/ https://drive.google.com/drive/u/0/folders/1xqek060j9_zVWGY80m33qSCrZf3scqUy
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