RuanPetrus/manga-text-extractor

Manga text extractor

Manga text extractor

This extract text from manga pages based on the paper.

Requirements

This project was created using python version 3.11. And the following dependencies

pip install numpy
pip install opencv-python
pip install matplotlib
pip install -U scikit-learn

Usage

The default model is the multilayer 13x13 berserk. So it only extract Berserk images.

In order to extract the text from a image you can run

python main.py extract <image-path>
python main.py extract ./source_images/berserk/6.jpg

Changing the model

The models are located in the models folder.
If you want to change the model you can change the following line in predict in the main file.

def predict(path: str) -> None:
    with open(MODELS_PATH + "<your-model>", "rb") as f:
        model = pickle.load(f)
    ...

Example:

def predict(path: str) -> None:
    with open(MODELS_PATH + "multilayer_13x13_onepiece", "rb") as f:
        model = pickle.load(f)
    ...

This will only works for models with 13x13 operators, if you want to change the size of operators you will have to change the following lines in the multilayer file:

...
OPERATOR_HEIGHT = 13
OPERATOR_WIDTH  = 13
...

TODO

• Neural nets?
• Support vector machines ?
• Better way to abstract and store models
• Refactor array code