This project is implemented using Tesseract OCR for character recognition in images and pdf.
Currently 4 languages are supported, English, Kannada, Tamil and Telegu.
A web interface is developed using Django framework, which allows users to upload an image or a pdf onto the webpage that returns the text grabbed from the file uploaded.
To install Django:
pip install Django
Install dependencies for ocr:
sudo apt-get install tesseract-ocr
pip install pytesseract
To work with images:
sudo pip install pillow
To work with pdf:
sudo apt-get install imagemagick
pip install wand
Other dependencies:
pip install autocorrect
To download the trained data for Kannada,Tamil and Telegu, go to
https://github.com/indic-ocr/tessdata/
The respective trained data will be found in the folders kan tam and tel. Download and copy those file into tessdata in your local folder where tesseract is installed.
At the terminal, go to the folder containing the project and type the following command to start the localhost server.
python manage.py runserver
To tag the few proper nouns in english that might not be recognised corectly, run the filtered.py script.
filtered.py depends on nltk tokenizing modules. To install them, run-
nltk.download('punkt')
The extraction is quite resource intensive and takes time(depends on the hardware used), especially if the uploaded file is a large pdf. It is advised to break the pdf into chunks of a couple of pages(8-10) per file and uploading them for quicker results.
The extracted text can also be found int the project folder in a text file extracted.txt.
Install the dependencies mentioned earlier and just download the ocr_on_terminal.py file. Copy the file(from which text is to be extracted) into the folder containing ocr_on_terminal_on_terminal_on_terminal.py.
Run ocr_on_terminal_on_terminal.py on the command prompt/terminal.
python ocr_on_terminal.py
Extracted text can also be found in extracted.txt inside the same directory.