SALAMI: Subjective Legibility Assessments for Ancient Manuscript Images

250 images of 50 manuscript regions annotated with spatial maps of mean legibility and uncertainty. The dataset serves as ground truth for developing quantitative evaluation metrics in digital text restoration. It was created by Simon Brenner at TU Wien based on assessments from 20 experts in philology and paleography.

Use Cases

• Train machine learning models for legibility prediction based on expert-provided spatial maps.
• Benchmark quantitative evaluation metrics for digital text restoration tasks.
• Analyze uncertainty in subjective legibility assessments across different manuscript images.
• Develop image processing algorithms for enhancing ancient manuscript readability.

Strengths

• Includes 250 annotated images derived from 50 distinct manuscript regions.
• Provides spatial maps of mean legibility and uncertainty (standard deviation).
• Ground truth is based on assessments from 20 domain experts in philology and paleography.
• Includes processing scripts in Python and R for reproducibility.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

TU Wien

Collection Method

Expert assessments from a study with 20 philologists and paleographers.