DemOCRatos - OCR for the people

Attempting to read portions of a screen or a video and exporting its data in a CSV file.

Blazingly slow, but it kinda works

Contents

1. Features
2. Install
3. Run
4. Notes
5. Dependencies

Features

• Real time recognitions of numbers of the screen.
• Offline number recognition in a video file.
• Works on easily configurable areas, as many as one wants.
• Integrated Tesseract OCR and EasyOCR.
• Easy to add new OCR methods (see src/ocr.py).
• Dead simple to use!

Install

This has been mainly developed and tested on Ubuntu 22.04, with Python 3.10.

# Install Python version >= 3.10 (necessary on lower Ubuntu versions)
sudo add-apt-repository ppa:deadsnakes/ppa
sudo apt update
sudo apt install python3.13 python3.13-venv python3.13-tk

# Install dependencies
sudo apt install tesseract-ocr libtesseract-dev

git clone ... && cd ocr_motor
python3.13 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

# Optional additional OCR methods
pip install easyocr

Run

source .venv/bin/activate
python gui.py

Notes

• Known OCR issues:
  • Characters confusion, depending on the font: 0 and 8, 1 and 7, 5 and 9.
  • Missing dot (e.g. 42.42 turned into 4242).
• Tips to improve OCR reliability:
  • Upscale the detected area, to get a better characters resolution.
  • Use min and max bounds to filter outliers out.
  • Don't trust the OCR output too much. Potentially implement post-filtering based on knowledge of the recorded data. For example if measuring a variable that can only evolve slowly, big jumps in the output value can be marked as outliers and discarded.
• When processing a video, enabling the preview can induce up to 20% overhead.
• EasyOCR requires PyTorch and Scipy, so isn't lightweight. The first time the program is started, it will download necessary model weights (stored in ~/.EasyOCR/model). See more details on the EasyOCR GitHub (link). With this application, it seems that EasyOCR is slower than Tesseract.

Dependencies

This work is merely a wrapper and a graphical interface for some already existing OCR implementations. It heavily uses Tkinter and CustomTkinter for the interface.

• OpenCV (MIT license): image processing.
• CustomTkinter (MIT license): beautiful interface and GUI.
• Tesseract (Apache 2.0): OCR API.
• Tesserocr (MIT license): Python wrapper for Tesseract.
• EasyOCR (Apache 2.0): another OCR API.

TODO (maybe one day)

• Loading and processing videos
• Saving/loading configuration
• Multi threading, for less blazing slowness
• Make it easier to add new OCR methods, and documenting it
• Logging not only in the Python terminal, but also in the logging text box
• Real time graphing