Kratos-AI/Korean-Signs-Image-Dataset

Korean Signs Image Dataset

This dataset contains a diverse collection of high-resolution images of Korean street signs, shop boards, and public information signs. The dataset has been curated and anonymized to support research in multilingual OCR, text detection, and cultural-linguistic visual understanding.

Contact

For queries or collaborations related to this dataset, contact:

• [email protected]
• [email protected]

Supported Tasks

• Task Categories:

  • Image Classification
  • Text Recognition (OCR)

• Supported Tasks:

  • Korean text detection and transcription in natural scenes
  • Multilingual OCR for signboards and public notices
  • Scene text localization and segmentation
  • Cross-lingual visual-language model training
  • Cultural and linguistic signage recognition in AI vision systems

Languages

• Primary Language: Korean
• Secondary Presence: Limited English or bilingual signage in urban areas

Dataset Creation

Curation Rationale

The dataset was created to advance OCR and text understanding capabilities in Korean, especially for multilingual environments. It enables researchers to train models for accurate text detection in real-world scenarios involving complex backgrounds, lighting conditions, and fonts.

Source Data

• Contributors: Volunteer photographers and crowd-sourced image collectors across major Korean cities
• Collection Process: Images of shop signs, street boards, warning signs, and public information displays were captured and manually filtered to remove identifiable people or private property indicators. All data was anonymized before release.

Other Known Limitations

• Bias: Urban areas may be overrepresented compared to rural regions
• Lighting Variability: Some images may include glare, reflections, or partial occlusion
• Cultural Specificity: Some signs may contain idiomatic or region-specific expressions that require contextual understanding

Intended Uses

✅ Direct Use

• Training OCR and computer vision models for Korean text
• Benchmarking multilingual text recognition systems
• Research on signage design and urban language patterns
• Developing accessibility tools for travelers and non-Korean speakers

❌ Out-of-Scope Use

• Identification or surveillance of individuals or locations
• Commercial use without proper license attribution
• Use in datasets involving sensitive or private content

License

CC BY 4.0