hygtfrde / delphi

Text extractor for video files

Delphī

An open sourced extractor for video files to raw text

---

Core Software Components in src

Page Extractor

• Utilize motion detection to recognize a full page, scanning its contents -- Extract the still frame
• Increment and count prefixes and page numbers

Text Extractor

• Feed in the raw pages as image files and generate text pages -- Utilizes OCR

Normalizer

• Detect edges and contours for accurate page detection
• Enlarge or shrink frame as necessary
• Flatten the image and remove curves

Quality Rating

• Input a frame and provide a quality score -- Ensuring high quality text extraction and confidence levels

---

Setting Up

Virtual Enviornment

• Start your Conda environment before downloading any modules with Pip
  
  • conda create --name delphi
  • conda activate delphi
  • Turn it off: conda deactivate
• For venv from Python Standard Library
  • Create a new Venv: python -m venv delphi
  • Windows: .\delphi\Scripts\activate
  • Mac/Linux: source delphi/bin/activate
  • Turn it off: deactivate

Installation

• Then use Pip to install dependencies
  pip install -r requirements.txt

---

Running

• Before extracting pages from a video, add your video file in videos directory.
• Match the name of the Video Path in main.py:

    # Video Path
    video_path = 'videos/test-scan.mov'

• Alternatively, set a custom Video Path for your liking
• Start program with: python main.py

Main.py

• Entry point to run program: python main.py
• Images will be extracted into a output_frames folder
• Text file will be saved as ..... WIP

Testing

Text Extraction

• To test text extraction from a still frame: python testing/test_main.py

Docker

• First, check services running and start your Docker Daemon: If using Homebrew:

brew services list
brew services start docker
brew services stop docker

If using MacOS Terminal:

launchctl start com.docker.dockerd
launchctl stop com.docker.dockerd

• For Keras and DocTR OCR programs, Docker is a preferred option
• Build a specific Docker image: docker build -t ocr_testing .
• Run the image and create a container: docker run -it ocr_testing
• Replace ocr_testing with another name for the tests if needed.
• Or just execute the Docker script and follow the prompts to build and run: sh run_docker.sh
• Inside of the /docker directory, docker_entrypoint.sh defines which tests to run, modify the test scripts if needed.