This is a basic OCR system that is created to help picking up text from video stream (or still images), and output JSON data on where and what was detected, it produces rect coordinates with text recognized and confidences.
In order to work this script requires that your machine has Python 3.6 or newer, OpenCV and Tesseract installed.
On Linux (and Mac?) these can be obtained from your OS package manager.
On Windows you have to manually download and install these libraries.
Before doing any setup grab OpenCV trained data
The link to the actual data may change, so open up this file in text editor and find the URL for 'EAST' model, download and save it.
https://github.com/opencv/opencv_extra/blob/master/testdata/dnn/download_models.py
MobileNet SSD v2 weights and config data https://github.com/opencv/opencv/wiki/TensorFlow-Object-Detection-API
Now let's move on to the platform specific packages.
Most systems should have everything required in their package management systems, on Ubuntu Linux for example this is all what you need
apt install libtesseract-dev tesseract-ocr-eng libopencv-dev
But also install python3 if not present yet.
apt install python3 pip3
Download and extract those in relevant location, later you will need to update system's %PATH% variable to point to OpenCV and Tesseract location
OpenCV https://github.com/opencv/opencv/releases
Tesseract 4.x https://github.com/UB-Mannheim/tesseract/wiki
You will need tesseract trained data for the required languages (copy this to tesseract/tessdata folder)
https://github.com/tesseract-ocr/tesseract/wiki/Data-Files
https://github.com/tesseract-ocr/tessdata
Additionally you will need prebuilt python package for tesseract (.whl file)
https://github.com/simonflueckiger/tesserocr-windows_build/releases
Now when you have those libraries let's start with creating virtual environment
Python 2.x (Linux) / Windows
python -m venv venv##
Python 3 (Linux)
python3 -m venv venv##
Where ## is your Python version, such as 37 for Python 3.7
Now we can activate it and install Python packages
Windows
./venv##/Scripts/activate
Linux
source ./venv##/bin/activate
And also update the newly created environment's pip
python -m pip install --upgrade pip
On Linux this script will set temporary alias so if previously you were using python3 explicitly now it will be just python. It also might add deactivate() shell function to clear this off so you can continue using your system Python without restarting the terminal
Remember that you will need to activate your environment every time you open new terminal, otherwise it'll use your system python and it will mess it up if you do pip install
Now we are ready to install python packages, let's do it
pip install -r requirements.txt
(Windows only) We have to install prebuilt .whl for tesserocr you've downloaded previously
pip install tesserocr-{downloaded.version}.whl
On some platforms we might have to update system's PATH environment variable, on Linux this may already handled when you did installation using package manager, on Windows however you have to do it manually
(PowerShell) set env vars (change to your paths)
$env:PATH+=";E:\tesseract"
$env:PATH+=";E:\opencv\build\x64\vc15\bin"
$env:TESSDATA_PREFIX+="E:\tesseract\tessdata"
And for reference here is what is expected
- $PATH: OpenCV .so/.dll, tesseract as well
- $TESSDATA_PREFIX: Should point to tesseract install dir tessdata/ subfolder, ensure it has language data
Everything should be ready to run the script.
Remember that you must activate the environment if not yet.
Finally let's run it.
python combined.py --input <path/to/file> [--width #] [--height #] [--frame #]
Use width and height to downscale images and improve performance You don't need to set width and height, however you should keep aspect ratio if you do, currently it must be multiple of 32, we'll change this later
You can also pass --frame <number> to pick specific frame from video file
Use --debug to see visualization
For testing purposes there is basic web server available, it accepts file with width and height using POST request.
To run this server grab flask using pip.
pip install flask
Follow flask instructions and set environment and flask script app(server.py without extension, in this case) to run, optionally using prod or development version
export FLASK_APP=server
export FLASK_ENV=development
Run the server.
flask run
Open http://127.0.0.1:5000 in a browser.
For more information refer to flask documentation.