Aadv1k / OctetOCR

Octet is an exploratory OCR or text recognition library to prepare and train upon raw data

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

OctetOCR

Octet is an OCR model training system, this includes functionality for both preparing and training the data.

This system implements a rudimentary K-NN function to predict the similarity of an OctetCharacter which contains the raw image data (unsigned char*) against a pre-computed array of characters derived from dataset/

NOTE Library under active development, things may change or break. I also appreciate any input on the code, cheers!

See:

Examples

See example.c for how you might use this library

Prepare

Loading a dataset from a folder is slow since everytime pre-processing has to be applied. Hence it is adviced that the data be serialized as a CSV and read from there instead, here is what that may look like :-

#include <octet.h>
  /* ... */

  OctetData* data;

  if (!dataFileExists) {
    data = octet_load_training_data_from_dir("./dataset");
    octet_write_training_data_to_csv("./data.csv");
  } else {
    data = octet_load_training_data_from_csv("./data.csv");
  }

  octet_free_training_data(data);

  /* ... */

Train

#include <octet.h>

  /* ... */

  OctetData* trainingData = octet_load_training_data_from_dir("./dataset");
  OctetCharacter testCharacter = octet_load_character_from_image("./tests/test_data/test-A.jpg");

  char predictedLabel = octet_k_nearest_neighbour(testCharacter, trainingData, /* k */ 3);
  assert(predictedLabel == 'A');

  octet_free_character(testCharacter);
  octet_free_training_data(trainingData);

  /* ... */

Gallery

The program has been tested against the following samples, feel free to add to the ./dataset, if you want an alt version of "A" in the dataset, jut name it something like A1.jpg

Input Training Data Match
A
B

Test

Win32

.\build.bat TEST
.\octet_test.exe

Unix

Credit to rarafael for the shell script

./build.sh TEST
./octet_test

Credits

Credit to rarafael for

  • Providing the build.sh script
  • Implementing octet_load_training_data_from_csv at src/prep.c
  • General quality of life improvements

About

Octet is an exploratory OCR or text recognition library to prepare and train upon raw data

License:GNU General Public License v3.0


Languages

Language:C 99.7%Language:Shell 0.2%Language:Batchfile 0.2%