This repository contains the fine-tuned Long Short-Term Memory (LSTM) model for Arabic text recognition in Tesseract OCR.
This project is part of a research study titled "Enhancing Arabic Text Recognition: Fine-tuning of the LSTM Model in Tesseract OCR". The LSTM model in Tesseract OCR was fine-tuned using a diverse training dataset of 1038 unique Arabic fonts. The performance of this fine-tuned model was evaluated across various Arabic text types and compared with the original Tesseract OCR model.
The ara.traineddata file in this repository is the result of our fine-tuning process. It is designed to significantly enhance Arabic text recognition in Tesseract OCR.
To use this fine-tuned model, download the ara.traineddata file and place it in your Tesseract 'tessdata' directory, replacing the existing Arabic trained data file. Then, simply run Tesseract as you normally would.
The fine-tuned model demonstrated significant performance enhancements in recognizing most Arabic text types, specifically Arabic-indic digits, European digits, normal Arabic text, and Arabic text without diacritics. For more details about the research and results, refer to the research paper.