shiima/Preprocessing-khalidchawtany-ckb_simple_ocr_dataset

CKB Simple OCR Dataset (Processed)

Dataset Description

This dataset contains 15,000 Central Kurdish (Sorani) OCR samples from the khalidchawtany/ckb_simple_ocr_dataset, preprocessed using the asosoft library for Kurdish text normalization.

Languages

• Central Kurdish (ckb) / Sorani Kurdish

Dataset Structure

The dataset maintains the same structure as the original CKB Simple OCR dataset with the following columns:

• image
• text

Data Processing

Text Preprocessing

All text fields in the dataset were preprocessed using the asosoft library with the following pipeline:

1. Text normalization (asosoft.Normalize) - Standardizes Kurdish characters and encoding
2. Punctuation normalization (asosoft.NormalizePunctuations) - Standardizes punctuation marks
3. Line trimming (asosoft.TrimLine) - Removes extra whitespace

This preprocessing ensures:

• Consistent Kurdish character encoding
• Standardized punctuation across the dataset
• Removal of extra whitespace
• Improved text quality for OCR training and evaluation
• Better compatibility with Kurdish NLP tools

Example Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("shiima/Preprocessing-khalidchawtany-ckb_simple_ocr_dataset")

# View a sample
sample = dataset['train'][0]
print("Sample:", sample)

# Access specific fields (adjust based on actual column names)
for sample in dataset['train']:
    # Process your OCR data here
    print(sample)
    break

Use Cases

This dataset is suitable for:

• Kurdish OCR model training
• Central Kurdish text recognition
• Handwritten Kurdish text recognition
• Kurdish document digitization
• OCR evaluation and benchmarking
• Kurdish language model fine-tuning
• Multi-script OCR systems

Source Dataset

This dataset is derived from:

• Original Dataset: khalidchawtany/ckb_simple_ocr_dataset
• Processing: All text fields preprocessed with asosoft
• No sampling: Complete dataset with all 15,000 samples

Statistics

• Total Samples: 15,000
• Language: Central Kurdish (Sorani)
• Script: Arabic script
• Format: Same structure as original dataset
• Columns: 2
• Preprocessing: asosoft normalization applied to all text fields

Quality Improvements

The asosoft preprocessing provides several quality improvements:

1. Character Standardization: Converts various Kurdish character encodings to a standard form
2. Punctuation Consistency: Normalizes different punctuation styles
3. Whitespace Handling: Removes unnecessary spaces and line breaks
4. Encoding Issues: Fixes common encoding problems in Kurdish text

These improvements make the dataset more suitable for:

• Training robust OCR models
• Consistent evaluation metrics
• Better generalization across different Kurdish text sources

Citation

If you use this dataset, please cite the original CKB Simple OCR dataset:

@misc{ckb_simple_ocr,
  title={CKB Simple OCR Dataset},
  author={Khalid Chawtany},
  year={2024},
  url={https://huggingface.co/datasets/khalidchawtany/ckb_simple_ocr_dataset}
}

License

This dataset inherits the MIT license from the original CKB Simple OCR dataset.

Acknowledgments

• Original dataset creator: Khalid Chawtany
• Preprocessing library: asosoft by Kurdish NLP
• Kurdish NLP community

Notes

• All text fields have been preprocessed for consistency
• The dataset structure remains identical to the original
• Image data (if present) is preserved without modification
• Suitable for training production-ready OCR models
• Compatible with popular OCR frameworks (Tesseract, EasyOCR, etc.)

Contact

For questions or issues, please open an issue on the dataset repository.