Arabic_OCR

This repo contains Arabic OCR App. The APP can be used to extract the Arabic text from the images. This was built based on the EasyOCR library. EsayOCR built detection/recognition model to detect and recognize the characters and words. For detection part they used the pretrained model for CRAFT algorithm. For recognition they built a CRNN model. For our case, we used the two pretrained model for Arabic language. To create the wep app, we used the Streamlit library.

App

The app deployed by Streamlit share. But the streamlit share dose not provide gpu at the backend only cpu. The app can run on cpu but with slow performance. So, the text extraction process can take from 2 to 3 minutes on cpu, extracting text from the same image with gpu in backend can lead to a huge improvement in the performance and the process can take 13 sec at average. So, we need to take it into consideration when trying it.

You can try the app on the streamlit sharing cloud on this link Arabic OCR

Installation

1. Install via conda

In this step we assume that conda is preinstalled on the machine. If conda is not installed you can follow the steps on the that link

1. At first we need to clone the repo to the local machine.

git clone https://github.com/maidaly/Arabic_OCR.git

2. Create a new conda enviroment to run the app inside it.

conda create --name arabic_ocr

conda activate arabic_ocr

3. Install the required python packages

pip install -r requirements.txt

4. Run the app

streamlit run app.py

The command need to run from the folder that contains the repo files. It will generate two links you can go throgh http://localhost:8501 to run the app on the local host.

2. Run via docker container

In this step we assume that Docker is running on your machine.

1. Clone the repo silmilar to conda installtion.
2. Convert directory to the repo location.
3. Build a docker image.

docker image build -t arabic_ocr:app 

4. Run the image

docker run -p 8501:8501 arabic_ocr:app 

After running the image we can go to http://localhost:8501 to run the app.

Note

• The first time running the app it may take time (some moments) to download the pretrained models that used. The time depends on the network speed. Then the pretrained models will be saved to used later.
• The app is running faster with the machine that contains Nvidia gpus. If the gpu is not availble the app will run but with slow performance.