This project implements OCR for television news from Bengali and Hindi news channels. I am using OpenCV along with a pre-trained tensorflow model called EAST(An Efficient and Accurte Scene Test detector) for detecting ROI (Regions of interest) from the news videos. Then the detected ROIs are extracted and OCR, implemented using tesseract 4.0 is used to exract the text.
apt update
apt install -y python3-pip build-essential libssl-dev libffi-dev python3-dev
apt install -y tesseract-ocr
apt install -y libtesseract-dev libsm-dev
pip3 install pytesseract opencv-python numpy
python3 textdetection.py <filename>
Specify additional argument 'O' for running OCR.
Output is saved to the file output/output.txt