Toolbox for OCR post-correction

Related tags

Computer Visionochre
Overview

Ochre

Ochre is a toolbox for OCR post-correction. Please note that this software is experimental and very much a work in progress!

  • Overview of OCR post-correction data sets
  • Preprocess data sets
  • Train character-based language models/LSTMs for OCR post-correction
  • Do the post-correction
  • Assess the performance of OCR post-correction
  • Analyze OCR errors

Ochre contains ready-to-use data processing workflows (based on CWL). The software also allows you to create your own (OCR post-correction related) workflows. Examples of how to create these can be found in the notebooks directory (to be able to use those, make sure you have Jupyter Notebooks installed). This directory also contains notebooks that show how results can be analyzed and visualized.

Data sets

Installation

git clone [email protected]:KBNLresearch/ochre.git
cd ochre
pip install -r requirements.txt
python setup.py develop
  • Using the CWL workflows requires (the development version of) nlppln and its requirements (see installation guidelines).
  • To run a CWL workflow type: cwltool|cwl-runner path/to/workflow.cwl <inputs> (if you run the command without inputs, the tool will tell you about what inputs are required and how to specify them). For more information on running CWL workflows, have a look at the nlppln documentation. This is especially relevant for Windows users.
  • Please note that some of the CWL workflows contain absolute paths, if you want to use them on your own machine, regenerate them using the associated Jupyter Notebooks.

Preprocessing

The software needs the data in the following formats:

  • ocr: text files containing the ocr-ed text, one file per unit (article, page, book, etc.)
  • gs: text files containing the gold standard (correct) text, one file per unit (article, page, book, etc.)
  • aligned: json files containing aligned character sequences:
{
    "ocr": ["E", "x", "a", "m", "p", "", "c"],
    "gs": ["E", "x", "a", "m", "p", "l", "e"]
}

Corresponding files in these directories should have the same name (or at least the same prefix), for example:

├── gs
│   ├── 1.txt
│   ├── 2.txt
│   └── 3.txt
├── ocr
│   ├── 1.txt
│   ├── 2.txt
│   └── 3.txt
└── aligned
    ├── 1.json
    ├── 2.json
    └── 3.json

To create data in these formats, CWL workflows are available. First run a preprocess workflow to create the gs and ocr directories containing the expected files. Next run an align workflow to create the align directory.

To create the alignments, run one of:

  • align-dir-pack.cwl to align all files in the gs and ocr directories
  • align-test-files-pack.cwl to align the test files in a data division

These workflows can be run as stand-alone; associated notebook align-workflow.ipynb.

Training networks for OCR post-correction

First, you need to divide the data into a train, validation and test set:

python -m ochre.create_data_division /path/to/aligned

The result of this command is a json file containing lists of file names, for example:

{
    "train": ["1.json", "2.json", "3.json", "4.json", "5.json", ...],
    "test": ["6.json", ...],
    "val": ["7.json", ...]
}
  • Script: lstm_synched.py

OCR post-correction

If you trained a model, you can use it to correct OCR text using the lstm_synced_correct_ocr command:

python -m ochre.lstm_synced_correct_ocr /path/to/keras/model/file /path/to/text/file/containing/the/characters/in/the/training/data /path/to/ocr/text/file

or

cwltool /path/to/ochre/cwl/lstm_synced_correct_ocr.cwl --charset /path/to/text/file/containing/the/characters/in/the/training/data --model /path/to/keras/model/file --txt /path/to/ocr/text/file

The command creates a text file containing the corrected text.

To generate corrected text for the test files of a dataset, do:

cwltool /path/to/ochre/cwl/post_correct_test_files.cwl --charset /path/to/text/file/containing/the/characters/in/the/training/data --model /path/to/keras/model/file --datadivision /path/to/data/division --in_dir /path/to/directory/with/ocr/text/files

To run it for a directory of text files, use:

cwltool /path/to/ochre/cwl/post_correct_dir.cwl --charset /path/to/text/file/containing/the/characters/in/the/training/data --model /path/to/keras/model/file --in_dir /path/to/directory/with/ocr/text/files

(these CWL workflows can be run as stand-alone; associated notebook post_correction_workflows.ipynb)

  • Explain merging of predictions

Performance

To calculate performance of the OCR (post-correction), the external tool ocrevalUAtion is used. More information about this tool can be found on the website and wiki.

Two workflows are available for calculating performance. The first calculates performance for all files in a directory. To use it type:

cwltool /path/to/ochre/cwl/ocrevaluation-performance-wf-pack.cwl#main --gt /path/to/dir/containing/the/gold/standard/ --ocr /path/to/dir/containing/ocr/texts/ [--out_name name-of-output-file.csv]

The second calculates performance for all files in the test set:

cwltool /path/to/ochre/cwl/ocrevaluation-performance-test-files-wf-pack.cwl --datadivision /path/to/datadivision.json --gt /path/to/dir/containing/the/gold/standard/ --ocr /path/to/dir/containing/ocr/texts/ [--out_name name-of-output-file.csv]

Both of these workflows are stand-alone (packed). The corresponding Jupyter notebook is ocr-evaluation-workflow.ipynb.

To use the ocrevalUAtion tool in your workflows, you have to add it to the WorkflowGenerator's steps library:

wf.load(step_file='https://raw.githubusercontent.com/nlppln/ocrevaluation-docker/master/ocrevaluation.cwl')
  • TODO: explain how to calculate performance with ignore case (or use lowercase-directory.cwl)

OCR error analysis

Different types of OCR errors exist, e.g., structural vs. random mistakes. OCR post-correction methods may be suitable for fixing different types of errors. Therefore, it is useful to gain insight into what types of OCR errors occur. We chose to approach this problem on the word level. In order to be able to compare OCR errors on the word level, words in the OCR text and gold standard text need to be mapped. CWL workflows are available to do this. To create word mappings for the test files of a dataset, use:

cwltool  /path/to/ochre/cwl/word-mapping-test-files.cwl --data_div /path/to/datadivision --gs_dir /path/to/directory/containing/the/gold/standard/texts --ocr_dir /path/to/directory/containing/the/ocr/texts/ --wm_name name-of-the-output-file.csv

To create word mappings for two directories of files, do:

cwltool  /path/to/ochre/cwl/word-mapping-wf.cwl --gs_dir /path/to/directory/containing/the/gold/standard/texts/ --ocr_dir /path/to/directory/containing/the/ocr/texts/ --wm_name name-of-the-output-file.csv

(These workflows can be regenerated using the notebook word-mapping-workflow.ipynb.)

The result is a csv-file containing mapped words. The first column contains a word id, the second column the gold standard text and the third column contains the OCR text of the word:

,gs,ocr
0,Hello,Hcllo
1,World,World
2,!,.

This csv file can be used to analyze the errors. See notebooks/categorize errors based on word mappings.ipynb for an example.

We use heuristics to categorize the following types of errors (ochre/ocrerrors.py):

  • TODO: add error types

OCR quality measure

Jupyter notebook

  • better (more balanced) training data is needed.

Generating training data

  • Scramble gold standard text

Ideas

  • Visualization of probabilities for each character (do the ocr mistakes have lower probability?) (probability=color)

License

Copyright (c) 2017-2018, Koninklijke Bibliotheek, Netherlands eScience Center

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Owner
National Library of the Netherlands / Research
National Library of the Netherlands / Research
National Library of the Netherlands / Research
Document manipulation detection with python

image manipulation detection task: -- tianchi function image segmentation salie

JiaKui Hu 3 Aug 22, 2022
OCRmyPDF adds an OCR text layer to scanned PDF files, allowing them to be searched

OCRmyPDF adds an OCR text layer to scanned PDF files, allowing them to be searched or copy-pasted. ocrmypdf # it's a scriptable c

jbarlow83 7.9k Jan 03, 2023
A toolbox of scene text detection and recognition

FudanOCR This toolbox contains the implementations of the following papers: Scene Text Telescope: Text-Focused Scene Image Super-Resolution [Chen et a

FudanVIC Team 170 Dec 26, 2022
Distilling Knowledge via Knowledge Review, CVPR 2021

ReviewKD Distilling Knowledge via Knowledge Review Pengguang Chen, Shu Liu, Hengshuang Zhao, Jiaya Jia This project provides an implementation for the

DV Lab 194 Dec 28, 2022
Code for the head detector (HeadHunter) proposed in our CVPR 2021 paper Tracking Pedestrian Heads in Dense Crowd.

Head Detector Code for the head detector (HeadHunter) proposed in our CVPR 2021 paper Tracking Pedestrian Heads in Dense Crowd. The head_detection mod

Ramana Subramanyam 76 Dec 06, 2022
Text page dewarping using a "cubic sheet" model

page_dewarp Page dewarping and thresholding using a "cubic sheet" model - see full writeup at https://mzucker.github.io/2016/08/15/page-dewarping.html

Matt Zucker 1.2k Dec 29, 2022
Crop regions in napari manually

napari-crop Crop regions in napari manually Usage Create a new shapes layer to annotate the region you would like to crop: Use the rectangle tool to a

Robert Haase 4 Sep 29, 2022
基于图像识别的开源RPA工具,理论上可以支持所有windows软件和网页的自动化

SimpleRPA 基于图像识别的开源RPA工具,理论上可以支持所有windows软件和网页的自动化 简介 SimpleRPA是一款python语言编写的开源RPA工具(桌面自动控制工具),用户可以通过配置yaml格式的文件,来实现桌面软件的自动化控制,简化繁杂重复的工作,比如运营人员给用户发消息,

Song Hui 7 Jun 26, 2022
The CIS OCR PostCorrectionTool

The CIS OCR Post Correction Tool PoCoTo Source code for the Java-based PoCoTo client enabling fast interactive batch corrections of complete OCR error

CIS OCR Group 36 Dec 15, 2022
A Tensorflow model for text recognition (CNN + seq2seq with visual attention) available as a Python package and compatible with Google Cloud ML Engine.

Attention-based OCR Visual attention-based OCR model for image recognition with additional tools for creating TFRecords datasets and exporting the tra

Ed Medvedev 933 Dec 29, 2022
Binarize document images

Binarization Binarization for document images Examples Introduction This tool performs document image binarization (i.e. transform colour/grayscale to

QURATOR-SPK 48 Jan 02, 2023
Links to awesome OCR projects

Awesome OCR This list contains links to great software tools and libraries and literature related to Optical Character Recognition (OCR). Contribution

Konstantin Baierer 2.2k Jan 02, 2023
Programa que viabiliza a OCR (Optical Character Reading - leitura óptica de caracteres) de um PDF.

Este programa tem o intuito de ser um modificador de arquivos PDF. Os arquivos PDFs podem ser 3: PDFs verdadeiros - em que podem ser selecionados o ti

Daniel Soares Saldanha 2 Oct 11, 2021
Assignment work with webcam

work with webcam : Press key 1 to use emojy on your face Press key 2 to use lip and eye on your face Press key 3 to checkered your face Press key 4 to

Hanane Kheirandish 2 May 31, 2022
Python Computer Vision application that allows users to draw/erase on the screen using their webcam.

CV-Virtual-WhiteBoard The Virtual WhiteBoard is a project I made using the OpenCV and Mediapipe Python libraries. Using your index and middle finger y

Stephen Wang 1 Jan 07, 2022
Code for AAAI 2021 paper: Sequential End-to-end Network for Efficient Person Search

This repository hosts the source code of our paper: [AAAI 2021]Sequential End-to-end Network for Efficient Person Search. SeqNet achieves the state-of

Zj Li 218 Dec 31, 2022
PSENet - Shape Robust Text Detection with Progressive Scale Expansion Network.

News Python3 implementations of PSENet [1], PAN [2] and PAN++ [3] are released at https://github.com/whai362/pan_pp.pytorch. [1] W. Wang, E. Xie, X. L

1.1k Dec 24, 2022
A fastai/PyTorch package for unpaired image-to-image translation.

Unpaired image-to-image translation A fastai/PyTorch package for unpaired image-to-image translation currently with CycleGAN implementation. This is a

Tanishq Abraham 120 Dec 02, 2022
BD-ALL-DIGIT - This Is Bangladeshi All Sim Cloner Tools

BANGLADESHI ALL SIM CLONER TOOLS INSTALL TOOL ON TERMUX $ apt update $ apt upgra

MAHADI HASAN AFRIDI 2 Jan 19, 2022
A Screen Translator/OCR Translator made by using Python and Tesseract, the user interface are made using Tkinter. All code written in python.

About An OCR translator tool. Made by me by utilizing Tesseract, compiled to .exe using pyinstaller. I made this program to learn more about python. I

Fauzan F A 41 Dec 30, 2022