OCR

Optical character recognition software written in C.

Requirements

• gcc
• GTK3
• SDL2
• SDL2_image

If you plan to train the network you will also need:

• python
• xelatex
• pdftoppm

Build

make

Usage

./ocr

Training

1. Generate the custom dataset used to train the neural network

   cd dataset
   # This can take a while
   ./generate_dataset.sh

2. (optional) Adjust training parameters in src/ocr_train.c
3. Use the --train option when launching the OCR

   ./ocr --train

This will output the neural network in output/ocr_network_eX after each epoch.

You can save the pre-computed dataset to avoid wasting time before each training (see details about this in src/ocr_train.c).