A vast dataset for Kurdish handwritten digits and isolated characters recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Natural Language Processing

Background:

• Developing robust optical character recognition (OCR) systems requires extensive datasets for diverse scripts.
• The Kurdish language, particularly its central dialect, lacks comprehensive digital resources for handwriting recognition.

Purpose of the Study:

• To introduce and describe two novel, large-scale datasets for Kurdish handwriting recognition: K-ZHMARA for digits and K-PIT for characters.
• To facilitate research and development of OCR technologies for the Kurdish language.

Main Methods:

• Data collection involved printing Kurdish digits and characters on A4 papers with grids.
• Image processing techniques including scanning, segmentation, cropping, resizing, binarization, and inversion were applied using Python.
• The K-ZHMARA dataset comprises 70,000 images of Kurdish digits (7,000 per digit).
• The K-PIT dataset includes 245,000 images of Kurdish characters (7,000 per character).

Main Results:

• The creation of two substantial datasets, K-ZHMARA and K-PIT, totaling 315,000 images.
• Standardized image processing pipelines were established for data preparation.

Conclusions:

• The presented datasets are valuable resources for training machine learning models for Kurdish handwriting recognition.
• These datasets are expected to significantly contribute to the advancement of Kurdish OCR technology and digital inclusion.