A curated dataset of paired handwritten exam answers and their transcribed text, designed for training optical character recognition (OCR) models on educational documents.

Usage Examples

Basic Dataset Loading

from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("JunaidMB/handwriting-ocr-images-dataset")

# Access specific splits
train_data = dataset['train']  # 62 samples
test_data = dataset['test']    # 16 samples

# View a single example
example = train_data[0]
print(f"Image size: {example['image_path'].size}")
print(f"Image mode: {example['image_path'].mode}")
print(f"Answer text: {example['answer_text']}")

Iterating Through Samples

from datasets import load_dataset

dataset = load_dataset("JunaidMB/handwriting-ocr-images-dataset")

# Iterate through training data
for sample in dataset['train']:
    image = sample['image_path']        # PIL.Image object
    answer_text = sample['answer_text'] # String

    # Your OCR model training code here
    # Example: Extract features, prepare for model
    print(f"Image: {image.size}, Text length: {len(answer_text)}")

Dataset Summary

This dataset contains 78 image-text pairs extracted from GCSE (General Certificate of Secondary Education) exam papers. Each sample consists of a cropped image of handwritten student work paired with the corresponding transcribed answer text from an exam grading database.

Key characteristics:

Images: Cropped JPEG images of handwritten regions (RGB, 200 DPI, variable sizes)
Text: Transcribed student answers (50-500+ characters each)
Source: UK secondary school GCSE exams (ages 14-16)
Subjects: Computer Science, History
Exam boards: OCR, Edexcel
Quality: Manually annotated bounding boxes, authenticated answer text from grading database

Supported Tasks and Leaderboards

This dataset is suitable for:

Optical Character Recognition (OCR) - Handwriting recognition
Handwriting Recognition - Character and word-level prediction
Document Understanding - Answer extraction and analysis
Educational Technology - Automated grading system development

Languages

English (UK educational context)
Academic writing style
Subject-specific terminology (Computer Science, History)

Dataset Structure

Data Instances

{
    'image_path': <PIL.Image.Image object>,
    'answer_text': 'The CPU (Central Processing Unit) is the main component...'
}

Data Fields

Field	Type	Description
`image_path`	`PIL.Image.Image`	Cropped JPEG image of handwritten answer region. RGB color space, variable size, 200 DPI resolution. Handwriting extracted via manual bounding box annotation.
`answer_text`	`string`	Student's answer text transcribed from handwriting. Plain text format with no markup. Sourced from exam grading database. Length varies 50-500+ characters.

Data Splits

Split	Examples	Purpose
Train	62	Model training
Test	16	Model evaluation

Split method: Fixed 80/20 stratified random split with seed=42 for reproducibility.

Dataset Creation

Curation Rationale

This dataset was created to support:

Educational Technology Research - Develop systems that can automatically extract and analyze student work
Handwriting Recognition - Build OCR models trained on authentic student handwriting, not synthetic data
Automated Assessment - Enable fair, consistent grading systems for educational institutions
Accessibility - Create better tools for students with different learning needs

The dataset focuses on GCSE exams to provide:

Real-world handwriting samples with varying legibility
Authentic academic content from multiple disciplines
Structured data for rigorous model evaluation

Personal and Sensitive Information

Privacy Considerations

⚠️ Important: This dataset contains student handwriting and academic work and should be handled with appropriate care.

Potential identifiers:

Handwriting style and characteristics (could be used to identify individuals)
Subject-specific knowledge and answer patterns
Temporal information (exam year implicit in curriculum)

Mitigation Measures:

✅ No student names, IDs, or personal identifiers included
✅ No student metadata beyond answer text
✅ Only answer content preserved (no question text or student context)
✅ Aggregated across multiple students

Recommendation: Users should:

Use this dataset responsibly for educational research and model development
Not attempt to reverse-identify students from handwriting
Respect educational data privacy principles
Comply with local data protection regulations (GDPR in UK context)

Known Limitations

Small Dataset Size:
- Only 78 total samples (62 train, 16 test)
- Limited compared to large-scale OCR datasets (thousands+)
- May not capture full diversity of handwriting patterns
Subject-Specific Vocabulary:
- Content focused on Computer Science and History terminology
- May not generalize to other domains
- Specialized academic language not representative of general writing
Fixed Image Quality:
- All images processed at 200 DPI
- No variation in image quality or scanning conditions
- May not transfer to real-world scanning scenarios
Binary Split Only:
- No validation split provided
- Only train/test split available
- Hyperparameter tuning requires careful cross-validation strategy
Variable Image Sizes:
- Cropped images vary in dimensions
- Requires preprocessing/normalization for model training
- No standardization across samples

Additional Information

Dataset Curators

Created by: JunaidMB
Github Profile: https://github.com/JunaidMB/
Hosted by: Hugging Face

Licensing Information

This dataset is released under the MIT License.

Terms: You are free to use, modify, and distribute this dataset for any purpose (including commercial use) with attribution.

Important Note: While the code and annotations are MIT licensed, be aware that:

GCSE exam content is copyrighted by exam boards (OCR, Edexcel)
Student work is educational content
Use responsibly and in compliance with local data protection laws (GDPR for UK context)

Citation Information

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

@dataset{handwriting_gcse_ocr_2024,
  title={Handwritten GCSE Exam Answers Dataset},
  author={JunaidMB},
  year={2024},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/JunaidMB/handwriting-ocr-images-dataset}}
}

APA Format:

JunaidMB. (2024). Handwritten GCSE Exam Answers Dataset [Data set]. Hugging Face.
https://huggingface.co/datasets/JunaidMB/handwriting-ocr-images-dataset

Questions and Feedback

For questions, issues, or suggestions:

Discussion: https://huggingface.co/datasets/JunaidMB/handwriting-ocr-images-dataset/discussions

Last Updated: January 2026 Dataset Version: 1.0

Downloads last month: 4