artificial-brix / brix-ocr

Hello, this is Brix OCR, an open-source project where we try to build an OCR engine with the help of others as well as make a custom model.

BRIX-OCR

OCR engines have been developed into many kinds of domain-specific OCR applications, such as receipt OCR, invoice OCR, check OCR, legal billing document OCR. They can be used for: Data entry for business documents, e.g. Cheque, passport, invoice, bank statement and receipt. Automatic number plate recognition. This project focuses to build a OCR engine with the help of datasets and images.

Available libraries

Some of the avalable OCR engines are:

• tesseract-ocr By google, GITHUB REPO | Documentation
• keras-ocr, GITHUB REPO | Documentation
• EasyOCR by Jaided AI , GITHUB REPO | Documentation
• Text detection with MSER and SWT by @azmiozgen, GITHUB REPO
• TeOCR by Hugging Face , Overview
• docTR by mindee, GITHUB REPO | Documentation

Datasets:

Some of the available datasets for testing and training a OCR engine:

• list open dataset about ocr.
• Keras OCR datatset
• IAM Handwriting
• ICDAR 2003
• TextOCR
• FUNSD (Form Understanding in Noisy Scanned Documents)
• ST-VQA
• SciTSR
• TextCaps
• DocBank
• Kannada-MNIST
• MLe2e

Drive Link of some datasets for testing and training a OCR engine:

• SciTSR
• DDI-100
• Incidental Scene
• Born Digital
• FUNSD
• SVT

Research papers:

• Attention-based Extraction of Structured Information from Street View Imagery
• TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models
• Adapting the Tesseract Open Source OCR Engine for Multilingual OCR
• PP-OCRv2: Bag of Tricks for Ultra Lightweight OCR System
• LayoutReader: Pre-training of Text and Layout for Reading Order Detection
• PP-OCR: A Practical Ultra Lightweight OCR System
• Robustness Evaluation of Transformer-based Form Field Extractors via Form Attacks
• End-to-End Interpretation of the French Street Name Signs Dataset
• MMOCR: A Comprehensive Toolbox for Text Detection, Recognition and Understanding

TASKS to resolve:

• Task 0: As we need to train the custom model as well as the pretrained models so we need datasets,please add datasets links or download them inside a drive and make hyper link in the datasets sections in the readme.md and complete the Task 0.
• Task 1: There are three folders given Newspaper ,Posters and Sheets,go inside one folder, you can find a image there, as a sample, please find similar images only and push them inside perticuler folders, minimum 50 images inside a folder will be enough to make the dataset.
• Task 2: In this task you have to make a jupyter notebook and in that try to use some of the given libraries in the readme section and you have to test their output using the images in the test images,and contribute a jupyter notebook as a name like this: Name_of_the_contributer.ipynb.
• Task 3: This is the last step of the project, as you have tried all the libraries,make a custom model using the datasets and the take the help of the research papers as well as you mentor of the project,make a jupyter notebook and complete the Task 3.

How to contribute

Please don't push any commits in the main branch, in that case the PR will not be accepted,as there are 4 tasks, please join the discord server first to contribute and then comment under the respective issues and then fork the repo and start working. HAPPY CONTRIBUTING!!!

brix-ocr's Issues

TASK 1: Making the dataset of Newspaper,Posters and Sheets and add them to their respective folder

TASK 1 :Building the test dataset:

Add 50 images in each folder of Posters, Newspapers and Sheets to make a contribution. A sample for each of them has been given in their respective folders.

• About the task: For better understanding, please have a look here: TASK 1
• contribution : For making a contribution please have a look here: Contribute

Since a person can make different datasets, multiple contributors can participate in it and submit it as a contribution.

TASK 3:Make a custom model for OCR using the dataset given in the README file and test it on the test images

Make a jupyter notebook, in which you can write the codes to train the OCR model on the dataset given in the README file and test your model on the images given in the test_images folder.

• About the task: For better understanding, please have a look here: TASK 3
• contribution : For making a contribution please have look here: Contribute

TASK 0: Find the open-source training dataset for OCR engine training and add the drive link in the README file

Find the open-source available datasets for training the OCR engine (minimum 3) and add it to your drive and then add the drive link in the dataset section of the README file.

• About the task: For better understanding, please have a look here: TASK 0
• contribution : For making a contribution please have a look here: Contribute

Since a person can find different datasets, multiple contributors can participate in it and submit their links as a contribution.

TASK 2: Make predictions on test images using open-source libraries available for OCR

Make a jupyter notebook, in which you can write the codes to do OCR on the images given in the test_images folder

• About the task : For better understanding, please have a look here: TASK 2
• contribution : For making a contribution please have look here: Contribute

Since there are multiple algorithms for making the predictions, multiple contributors can participate in it and make their own notebook as a contribution.