End-to-End Transcript Alignment of 17th Century Manuscripts: The Case of Moccia Code

Area of Science:

• Digital Humanities
• Computer Vision
• Document Image Analysis

Background:

• Digital libraries increasingly house historical handwritten documents.
• Accurate word-level alignment between transcriptions and document images is crucial for scholarly study and automated processing.
• Existing methods may struggle with the complexities of historical scripts and document layouts.

Purpose of the Study:

• To propose a novel, learning-free method for automatic word-level text-to-image alignment in historical documents.
• To address challenges in segmenting text lines with curved baselines.
• To handle word-level under- and over-segmentation errors during alignment.

Main Methods:

• A learning-free approach for transcription-to-image alignment.
• A line-level segmentation algorithm designed for curved baselines.
• A text-to-image alignment algorithm robust to segmentation errors.

Main Results:

• The line segmentation algorithm achieved 92% accuracy on a 17th-century Italian manuscript.
• The text-to-image alignment achieved an accuracy greater than 68%.
• Performance favorably compares with state-of-the-art methods on widely used datasets.

Conclusions:

• The proposed method offers an effective solution for aligning transcriptions to historical document images.
• The approach is valuable for digital humanities research and automated document analysis.
• The learning-free nature makes it broadly applicable without extensive training data.