|
|
--- |
|
|
license: mit |
|
|
language: |
|
|
- tr |
|
|
tags: |
|
|
- ocr |
|
|
- turkish |
|
|
- synthetic |
|
|
- handwritten |
|
|
- computer-vision |
|
|
size_categories: |
|
|
- 1K<n<10K |
|
|
--- |
|
|
|
|
|
# Turkish OCR Synthetic Dataset |
|
|
|
|
|
A synthetic dataset for Turkish OCR (Optical Character Recognition) containing 1,000 realistic images with Turkish text. |
|
|
|
|
|
## Features |
|
|
- **Text types**: Handwritten (70%) and printed (30%) |
|
|
- **Full Turkish character support**: Ç, ç, Ğ, ğ, İ, ı, Ö, ö, Ş, ş, Ü, ü |
|
|
- **Realistic backgrounds**: Lined paper, grid paper, and textured paper |
|
|
- **Augmentations**: Motion blur, noise, perspective distortion, and ink artifacts |
|
|
|
|
|
## Data Format |
|
|
Each sample contains: |
|
|
- `image`: 640×384 RGB image with overlaid text |
|
|
- `text`: Ground truth transcription |
|
|
- `text_type`: `"handwritten"` or `"printed"` |
|
|
|
|
|
## Usage |
|
|
```python |
|
|
from datasets import load_dataset |
|
|
dataset = load_dataset("emredeveloper/turkish-ocr") |
