Handwritten Text Recognition (HTR) system implemented using TensorFlow 2.x and trained on the Bentham/IAM/Rimes/Saint Gall/Washington offline HTR datasets. This Neural Network model recognizes the text contained in the images of segmented texts lines.
Data partitioning (train, validation, test) was performed following the methodology of each dataset. The project implemented the HTRModel abstraction model (inspired by CTCModel) as a way to facilitate the development of HTR systems.
Notes:
- All references are commented in the code.
- This project doesn't offer post-processing, such as Statistical Language Model.
- Check out the presentation in the doc folder.
- For more information and demo run step by step, check out the tutorial on Google Colab/Drive.
Datasets supported
a. Bentham
b. IAM
c. Rimes
d. Saint Gall
e. Washington
Requirements
- Python 3.x
- OpenCV 4.x
- editdistance
- TensorFlow 2.x
Command line arguments
--source: dataset/model name (bentham, iam, rimes, saintgall, washington)--arch: network to be used (puigcerver, bluche, flor)--transform: transform dataset to the HDF5 file--cv2: visualize sample from transformed dataset--kaldi_assets: save all assets for use with kaldi--image: predict a single image with the source parameter--train: train model using the source argument--test: evaluate and predict model using the source argument--norm_accentuation: discard accentuation marks in the evaluation--norm_punctuation: discard punctuation marks in the evaluation--epochs: number of epochs--batch_size: number of the size of each batch
Tutorial (Google Colab/Drive)
A Jupyter Notebook is available to demo run, check out the tutorial on Google Colab/Drive.
Sample
Bentham sample with default parameters in the tutorial file.
- Preprocessed image (network input)
- TE_L: Ground Truth Text (label)
- TE_P: Predicted text (network output)
Citation
If this project helped in any way in your research work, feel free to cite the following papers.
HTR-Flor++: A Handwritten Text Recognition System Based on a Pipeline of Optical and Language Models (here)
This work aimed to propose a different pipeline for Handwritten Text Recognition (HTR) systems in post-processing, using two steps to correct the output text. The first step aimed to correct the text at the character level (using N-gram model). The second step had the objective of correcting the text at the word level (using a word frequency dictionary). The experiment was validated in the IAM dataset and compared to the best works proposed within this data scenario.
@inproceedings{10.1145/3395027.3419603,
author = {Neto, Arthur F. S. and Bezerra, Byron L. D. and Toselli, Alejandro H. and Lima, Estanislau B.},
title = {{HTR-Flor++:} A Handwritten Text Recognition System Based on a Pipeline of Optical and Language Models},
booktitle = {Proceedings of the ACM Symposium on Document Engineering 2020},
year = {2020},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
location = {Virtual Event, CA, USA},
series = {DocEng '20},
isbn = {9781450380003},
url = {https://doi.org/10.1145/3395027.3419603},
doi = {10.1145/3395027.3419603},
}
Towards the Natural Language Processing as Spelling Correction for Offline Handwritten Text Recognition Systems (here)
This work aimed a deep study within the research field of Natural Language Processing (NLP), and to bring its approaches to the research field of Handwritten Text Recognition (HTR). Thus, for the experiment and validation, we used 5 datasets (Bentham, IAM, RIMES, Saint Gall and Washington), 3 optical models (Bluche, Puigcerver, Flor), and 8 techniques for text correction in post-processing, including approaches statistics and neural networks, such as encoder-decoder models (seq2seq and Transformers).
@article{10.3390/app10217711,
author = {Neto, Arthur F. S. and Bezerra, Byron L. D. and Toselli, Alejandro H.},
title = {Towards the Natural Language Processing as Spelling Correction for Offline Handwritten Text Recognition Systems},
journal = {Applied Sciences},
pages = {1-29},
month = {10},
year = {2020},
volume = {10},
number = {21},
url = {https://doi.org/10.3390/app10217711},
doi = {10.3390/app10217711},
}
HDSR-Flor: A Robust End-to-End System to Solve the Handwritten Digit String Recognition Problem in Real Complex Scenarios (here)
This work aimed to propose the optical model for Handwritten Digit String Recognition (HDSR) and compare it with the state-of-the-art models. The International Conference on Frontiers of Handwriting Recognition (ICFHR) 2014 competition on HDSR were used as baselines toevaluate the effectiveness of our proposal, whose metrics, datasets and recognition methods were adopted for fair comparison. Furthermore, we also use a private dataset (Brazilian Bank Check - Courtesy Amount Recognition), and 11 different approaches from the state-of-the-art in HDSR, as well as 2 optical models from the state-of-the-art in Handwritten Text Recognition (HTR).
@article{10.1109/ACCESS.2020.3039003,
author = {Neto, Arthur F. S. and Bezerra, Byron L. D. and Lima, Estanislau B. and Toselli, Alejandro H.},
title = {{HDSR-Flor:} A Robust End-to-End System to Solve the Handwritten Digit String Recognition Problem in Real Complex Scenarios},
journal = {IEEE Access},
pages = {208543-208553},
month = {11},
year = {2020},
volume = {8},
isbn = {2169-3536},
url = {https://doi.org/10.1109/ACCESS.2020.3039003},
doi = {10.1109/ACCESS.2020.3039003},
}
HTR-Flor: A Deep Learning System for Offline Handwritten Text Recognition (here)
This work aimed to propose the optical model for Handwritten Text Recognition (HTR) and compare it with the state-of-the-art models. The performance comparison was validated in 5 different datasets (Bentham, IAM, RIMES, Saint Gall and Washington). In addition, it was considered one of the best papers in the 33rd SIBGRAPI (2020).
@inproceedings{10.1109/SIBGRAPI51738.2020.00016,
author = {Neto, Arthur F. S. and Bezerra, Byron L. D. and Toselli, Alejandro H. and Lima, Estanislau B.},
title = {{HTR-Flor:} A Deep Learning System for Offline Handwritten Text Recognition},
booktitle = {2020 33rd SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)},
pages = {54-61},
month = {11},
year = {2020},
location = {Recife/Porto de Galinhas, PE, Brazil},
series = {SIBGRAPI' 33},
publisher = {IEEE Computer Society},
address = {Los Alamitos, CA, USA},
url = {https://doi.org/10.1109/SIBGRAPI51738.2020.00016},
doi = {10.1109/SIBGRAPI51738.2020.00016},
}
933 Dec 29, 2022
2.4k Jan 8, 2023
3.2k Jan 4, 2023
3.3k Dec 30, 2022
496 Jan 5, 2023
354 Dec 12, 2022
684 Jan 6, 2023
1 Jan 8, 2022
147 Dec 20, 2022
5 Nov 24, 2021
4 Aug 11, 2022
117 Nov 10, 2022
1 Oct 28, 2021
67 Dec 29, 2022
48 Jan 02, 2023
235 Dec 18, 2022
13.2k Jan 02, 2023
19 Dec 07, 2022
648 Dec 30, 2022
7 Dec 27, 2021
45 Dec 30, 2022
3 Jul 13, 2022
671 Dec 27, 2022
224 Dec 28, 2022
1 Jan 02, 2022
20 Jan 06, 2023
6 Sep 29, 2022