Fraktur Script Text-Line Image Test Set

Impresso Project's Frakturline Testset is a balanced evaluation set of 2,000 scanned text-line images for binary classification. It contains 1,000 images each of Fraktur (blackletter) and other scripts, primarily Antiqua. The dataset was developed for the Impresso digital humanities project.

Use Cases

• Train a binary image classifier to distinguish Fraktur from other script using 2,000 labeled text-line images.
• Benchmark vision model performance on the specific task of Fraktur script recognition.
• Analyze model confusion between the 'fraktur' and 'other' classes to assess script discrimination accuracy.

Strengths

• Balanced class distribution with exactly 1,000 images per class.
• Specifically designed as a held-out evaluation set for a defined binary classification task.
• Contains 2,000 total images of scanned historical text-lines.

Limitations

• Limited to 2,000 images, which may be insufficient for training large models from scratch.
• No information provided on image resolution, source document quality, or potential variations within the 'other' script class.
• Dataset details like image format specifics are not fully described in the provided input.

Provenance

Source

Impresso Project

Collection Method

Scanned text-line images compiled for a digital humanities project.

Time Range

null

Freshness

null

Geography

null