Ninshiki OCR

A basic Optical Character Recognition program written in C for the third semester of EPITA.

Build

• git clone https://github.com/navadta/ninshiki
• make
• If you want to train a neural network with the ./ocr/build/example/ocr example or in the GUI, you will need to generate a dataset, for this, run ./ocr/build/example/generate_dataset ./resources/dataset <image> by replacing image with the 3 fonts images present in the dataset folder.
  • ./ocr/build/example/generate_dataset ./resources/dataset arial.png
  • ./ocr/build/example/generate_dataset ./resources/dataset nunito.png
  • ./ocr/build/example/generate_dataset ./resources/dataset roboto.png
• All of the executable demo files will be in ./ocr/build/example/
• The GUI executable file will be in ./gui/build/

Demos

• Images
  • This demo will produce a grayscaled and a binarized version of the input image
  • Several images are available in the ./resources path
  • Run ./ocr/build/example/images path image_file (./ocr/build/example/images ./resources snoopy.bmp)
• Segmentation
  • This demo will produce an image with characters, words and lines underlined
  • An example text image is available in ./resources/text.bmp
  • ./ocr/build/example/segmentation path image_file (./ocr/build/example/segmentation ./resources text.bmp)
• XOR Neural Network
  • This demo will train a neural network to learn the Bitwise XOR function then it will output some results
  • ./ocr/build/example/xor_network activation_function (./ocr/build/example/xor_network 1)
  • 1 to use the Sigmoid activation function and 2 for ELU (Exponential Linear Unit)
• Dataset Generation
  • This demo will generate a dataset as explained in the Build section
• OCR Training
  • This demo will train a neural network capable of recognizing characters

GUI

• This demo will open a GUI with multiple functionalities to recognize the text from an image
• Run the GUI with ./gui/build/gui

Navadta Members:

- Alan GUERET

- Theo DARONAT

- Adam THIBERT

- Louis D'HOLLANDE

License

MIT © Navadta