aabdelfattah / OCRA

OCR for Arabic Cursive Text

OCRA : OCR for Arabic Cursive Text

A very basic implementation of Arabic OCR using MATLAB, it uses HMM for learning and recognition.

To run the application

• Add all the folders and subfolders to MATLAB Path.
• From the current directory type : main in MATLAB cmd window

Parameters

• To change number of runs : modify NUM_RUNS in main.m
• To choose Sliding window technique pass 1 to OCRA2() in main.m
• Some parameters are available in OCRA2.m
  • TRAINING_SAMPLES = 20 ; % number of pictures used for training per symbol(letter)
  • EVAL_SAMPLES = 2; % number of pictures used for evaluation per symbol(letter)
  • NUM_LETTERS = 28; % number of symbols (letters)
  • NUM_STATES = 5; % number of states in a HMM symbol model

Application Hierarchy

• letter.m : LUT maps between symbol string and symbol number for convenience
• load_training_data.m : loads BOTH training and evaluation data for direction sequence technique
• load_training_data_sw.m : loads BOTH training and evaluation data for sliding window technique
• Sliding2.m : extracts sliding window feature vector
• OCRA2.m : OCR training, recognition and error evaluation
• main.m : top level module
• OCRA29062014.pdf : project documentation

Directory

• SW : contains sliding window training data and evaluation data(.png) in a subdir eval in each letter dir
• set n : contains direction sequence training data and evaluation data(.txt files) in a subdir eval in each letter dir

currently set 3 saad is used

Disclaimer

This implementation was a part of a graduate course project and actually it achieves very poor result, you can't rely on it for any real OCR but it will be useful as a starting point to learn from and build on.

Contribute

• Fork project
• Add features
• Run tests
• Send pull request

License

See LICENSE file for details