metadata

license: mit
task_categories:
  - object-detection
language:
  - he
tags:
  - hebrew
  - manuscripts
  - digital-humanities
size_categories:
  - n<1K

Hebrew Letter Recognition Dataset

Dataset Description

This dataset contains images of Hebrew letters and the stop symbols for training and evaluating optical character recognition (OCR) models. The dataset is designed to support the development of machine learning models capable of identifying individual Hebrew letters from images, making it ideal for tasks such as:

OCR for Hebrew texts
Educational tools for Hebrew language learners
Digitization of historical Hebrew manuscripts

Dataset Structure

The dataset is organized into directories, with each directory corresponding to a specific Hebrew letter or the stop symbol. Each directory contains several images (in .jpg format) of that letter or symbol in different fonts, sizes, and variations.

Directory Structure:

/dataset/
    /א/     (Images of the letter "Aleph")
    /ב/     (Images of the letter "Bet")
    /ג/     (Images of the letter "Gimel")
    ...
    /stop/  (Images of the stop symbol ".")

Total Classes: 29 (28 Hebrew letters + 1 stop symbol)
File Format: .jpg
Image Size: Varies, typically 64x64 pixels

Example Directory Structure:

dataset/
    א/
        1.jpg
        2.jpg
        ...
    ב/
        1.jpg
        2.jpg
        ...
    stop/
        1.jpg
        2.jpg
        ...

Class Labels:

The dataset includes the following classes (letters and stop symbol):
- א, ב, ג, ד, ה, ו, ז, ח, ט, י, ך, כ, ל, ם, מ, ן, נ, ס, ע, ף, פ, ץ, צ, ק, ר, ש, ת, stop (.)

Dataset Use Cases

This dataset can be used to train machine learning models for:

Hebrew Letter Recognition: Build a model that recognizes individual Hebrew letters from scanned documents or photos of text.
OCR Systems: Develop OCR systems that digitize printed or handwritten Hebrew documents.
Educational Tools: Create applications for teaching Hebrew reading by recognizing letters in real-time.

Data Preprocessing

Preprocessing Steps:

Resizing: All images should be resized to a consistent size (e.g., 64x64 pixels) for input into a CNN model.
Normalization: Normalize pixel values to the range [0, 1] by dividing by 255.
Augmentation (optional): Apply data augmentation techniques such as rotations, flips, and zooms to increase the robustness of the model.

Example of loading and preprocessing the dataset using TensorFlow:

from tensorflow.keras.preprocessing import image_dataset_from_directory

# Load the dataset from the directory
dataset = image_dataset_from_directory(
    'path_to_dataset_directory',
    image_size=(64, 64),
    batch_size=32,
    label_mode='categorical'  # Multiclass classification
)

Dataset Statistics:

Total Images: 307
Number of Classes: 29 (28 letters + 1 stop symbol)
Image Format: .jpg
Average Images per Class: Approximately 10-15 images per class.

License:

This dataset is provided under the MIT License. You are free to use, modify, and distribute the dataset as long as you include attribution to the original author.

Citation:

If you use this dataset in your research or work, please cite it as follows:

@misc{hebrew-letter-dataset,
  title={Hebrew Letter Recognition Dataset},
  author={Benjamin Schnabel},
  year={2024},
  howpublished={\url{https://huggingface.co/datasets/your-dataset}},
}

Contributions:

If you would like to contribute additional variations of Hebrew letters or improve the dataset, feel free to submit a pull request or open an issue.

This README provides users with a clear understanding of what your dataset contains, how to use it, and some basic information about preprocessing and structure.

Let me know if you'd like further customization or additional details!