README.md

# Text Recognition Dataset 
This dataset is a comprehensive collection of Arabic and English text recognition samples designed for benchmarking and evaluating text recognition models. The dataset combines multiple sources to provide diverse text recognition challenges across different domains, scripts, and image conditions.

---

## 📁 Dataset Structure
- Images are stored in .png format
- Accompanied by a CSV file (TextRecognitionDatasetCompleteAnnotation.csv) with ground truth annotations and text characteristics.

---

## 📚 Data Sources and Attribution

This dataset is compiled from samples taken from the following sources:

- **EvArEST Dataset**: Arabic scene text samples from [HGamal11/EvArEST-dataset-for-Arabic-scene-text](https://github.com/HGamal11/EvArEST-dataset-for-Arabic-scene-text)
- **AROCRbench_ISIPPT Dataset**: Arabic OCR benchmarking samples
- **OCR-Arabic-Dates Dataset**: Arabic date recognition samples
- **Synthetic Document Datasets**: Artificially generated text samples for various categories including:
  - English Personal Names, IDs, Numbers, Dates, Characters, Words, Sentences
  - Arabic Characters, Names, Numbers, Words
  - Hijri Dates and other specialized text types

*Note: This dataset contains selected samples from the above sources, with additional text characteristics and annotations added for comprehensive text recognition evaluation.*

---

## 📊 Dataset Overview
- **Total Samples**: 3,008
- **Number of Columns**: 11
- **File Size**: 2.02 MB
- **Languages**: Arabic, English.
- **Image Format**: `.png`

---

## 📊 Dataset Statistics

| Column | Unique Values | Missing Values | Missing % |
|--------|---------------|----------------|-----------|
| image_id | 3,008 | 0 | 0.0% |
| Ground_Truth_Text | 2,232 | 0 | 0.0% |
| dataset_name | 16 | 0 | 0.0% |
| image_category | 4 | 0 | 0.0% |
| language | 3 | 0 | 0.0% |
| type_of_word | 8 | 0 | 0.0% |
| text_type | 1 | 0 | 0.0% |
| augmentation_type | 27 | 0 | 0.0% |
| Environment | 2 | 0 | 0.0% |
| difficulty | 4 | 0 | 0.0% |
| text_annotation_level | 2 | 0 | 0.0% |

---

## 📊 Data Distribution

### Dataset Composition
| Dataset Source | Count | Percentage |
|----------------|-------|------------|
| evarest | 1,588 | 52.8% |
| hijri_dates | 100 | 3.3% |
| English Personal Names | 100 | 3.3% |
| English IDs | 100 | 3.3% |
| English Numbers | 100 | 3.3% |
| English Dates | 100 | 3.3% |
| arocrbench_isippt | 100 | 3.3% |
| English Characters | 100 | 3.3% |
| ocr-arabic-dates | 100 | 3.3% |
| arabic_sentence_images | 99 | 3.3% |
| Arabic Characters | 99 | 3.3% |
| Arabic Names | 99 | 3.3% |
| Arabic Numbers | 93 | 3.1% |
| English Words | 89 | 3.0% |
| English Sentences | 81 | 2.7% |
| Arabic-word | 60 | 2.0% |

### Language Distribution
| Language | Count | Percentage |
|----------|-------|------------|
| Arabic | 1,940 | 64.5% |
| English | 1,063 | 35.3% |
| Other | 5 | 0.2% |

### Image Categories
| Category | Count | Percentage |
|----------|-------|------------|
| scene_text | 1,588 | 52.8% |
| Synthetic Document | 1,160 | 38.6% |
| Document | 160 | 5.3% |
| egyptian_id | 100 | 3.3% |

---

## 📝 Column Descriptions

- **image_id**: 3,008 unique values (unique identifier for each image)

- **Ground_Truth_Text**: 2,232 unique values (actual text content in the images)

- **dataset_name**: 16 unique values representing different data sources
  - evarest: 1,588 (52.8%)
  - hijri_dates: 100 (3.3%)
  - English Personal Names: 100 (3.3%)
  - English IDs: 100 (3.3%)
  - English Numbers: 100 (3.3%)
  - English Dates: 100 (3.3%)
  - arocrbench_isippt: 100 (3.3%)
  - English Characters: 100 (3.3%)
  - ocr-arabic-dates: 100 (3.3%)
  - arabic_sentence_images: 99 (3.3%)
  - Arabic Characters: 99 (3.3%)
  - Arabic Names: 99 (3.3%)
  - Arabic Numbers: 93 (3.1%)
  - English Words: 89 (3.0%)
  - English Sentences: 81 (2.7%)
  - Arabic-word: 60 (2.0%)

- **image_category**: 4 unique values representing image types
  - scene_text: 1,588 (52.8%)
  - Synthetic Document: 1,160 (38.6%)
  - Document: 160 (5.3%)
  - egyptian_id: 100 (3.3%)

- **language**: 3 unique values
  - Arabic: 1,940 (64.5%)
  - English: 1,063 (35.3%)
  - Other: 5 (0.2%)

- **type_of_word**: 8 unique values representing text content types
  - String: 1,976 (65.7%)
  - Date: 300 (10.0%)
  - Number: 281 (9.3%)
  - Character: 219 (7.3%)
  - AlphaNumeric: 100 (3.3%)
  - Sentence: 99 (3.3%)
  - Mixed: 28 (0.9%)
  - Other: 5 (0.2%)

- **text_type**: 1 unique value (all samples are typed text)
  - Typed: 3,008 (100.0%)

- **augmentation_type**: 27 unique values representing image processing techniques
  - colored: 1,357 (45.1%)
  - original: 730 (24.3%)
  - heavy_salt_pepper: 53 (1.8%)
  - morph_close_bleed: 53 (1.8%)
  - bleed_pepper: 50 (1.7%)
  - heavy_blur_morph: 47 (1.6%)
  - light_salt: 47 (1.6%)
  - subtle_bleed: 45 (1.5%)
  - subtle_noise: 41 (1.4%)
  - light_closing: 41 (1.4%)
  - combo_light: 41 (1.4%)
  - light_opening: 41 (1.4%)
  - horizontal_blur: 39 (1.3%)
  - light_pepper: 37 (1.2%)
  - combined_subtle: 35 (1.2%)
  - blur_tiny_salt: 35 (1.2%)
  - morph_noise: 34 (1.1%)
  - light_erosion: 33 (1.1%)
  - light_blur_salt: 33 (1.1%)
  - multi_morph: 32 (1.1%)
  - morph_open_blur: 31 (1.0%)
  - light_dilation: 31 (1.0%)
  - vertical_blur: 29 (1.0%)
  - offset_bleed: 26 (0.9%)
  - offset_bleed_noise: 25 (0.8%)
  - salt_blur: 23 (0.8%)
  - light_blur: 19 (0.6%)

- **Environment**: 2 unique values indicating data processing state
  - augmented: 2,278 (75.7%)
  - clean: 730 (24.3%)

- **difficulty**: 4 unique values representing recognition complexity
  - scene text: 1,588 (52.8%)
  - medium: 775 (25.8%)
  - hard: 345 (11.5%)
  - easy: 300 (10.0%)

- **text_annotation_level**: 2 unique values indicating annotation granularity
  - word level: 2,529 (84.1%)
  - sentence level: 479 (15.9%)

---

## 🛠️ How to Use
You can load the dataset using:

```python
from datasets import load_dataset

ds = load_dataset("BatSilver/TextRecognition_Document_RealScene_Data")
```

---

## 🎯 Use Cases
This dataset is ideal for:
- Text recognition model training and evaluation
- Cross-lingual OCR system development
- Multi-domain text recognition challenges