Tesseract OCR of Pokemon (or video game) dialog text on streaming video (modern version).
This project reads streaming video and runs OCR on defined regions for streams such as TwitchPlaysPokemon. For previous information, see version 1 of this project.
Note: work in progress
- Tesseract 4
- tessdata_best or tessdata_fast
- Which should contain: eng jpn chi_sim chi_tra kor spa deu ita
- Leptonica
- ffmpeg
- OpenCV 4
- EAST trained model
- Download the Tensorflow trained model from this link
Compiling:
- C++17 compiler and associated C++ runtime
- CMake 1.15 or newer
- tomlplusplus headers
mkdir -p build
cd build
cmake .. -D CMAKE_BUILD_TYPE=Release
cmake --build . --config Release
Optional:
cmake --install --config Release --prefix install_prefix
Basic usage:
./build/tppocr --help
./build/tppocr CONFIG_FILE URL_OR_FILE_PATH
Example:
./build/tppocr --debug-window --frame-stepping data/tpp-sword-720p.toml sample_images/sword_720p_narrator_dialog.png