A benchmark dataset for text line segmentation in palm leaf documents.

Area of Science:

• Digital Humanities
• Information Science
• Cultural Heritage Preservation

Background:

• Palm leaf manuscripts represent a significant historical writing medium in South and Southeast Asia, containing vast knowledge across disciplines.
• These invaluable manuscripts are susceptible to degradation from aging and environmental factors, necessitating urgent preservation efforts.
• Digitization is crucial for safeguarding and disseminating the knowledge contained within these ancient texts.

Purpose of the Study:

• To address the lack of diverse datasets for palm leaf manuscript digitization.
• To improve text line segmentation and recognition for enhanced digitization performance.
• To introduce a valuable resource for researchers in the field of historical document analysis.

Main Methods:

• Development of LeafOCR-Line, a novel dataset comprising 1710 masked palm leaf manuscript text lines.
• Inclusion of deterioration level annotations for each manuscript within the dataset.
• Conducting extensive experiments to validate the dataset's quality and utility.

Main Results:

• The LeafOCR-Line dataset provides a diverse and annotated resource for palm leaf text line segmentation.
• Experimental results confirm the high quality and practical applicability of the proposed dataset.
• The dataset facilitates research into more robust and accurate digitization of historical manuscripts.

Conclusions:

• LeafOCR-Line is a significant contribution to the field of palm leaf manuscript digitization.
• The dataset will serve as a valuable public resource for advancing research in text line segmentation and recognition.
• Availability of LeafOCR-Line is expected to accelerate progress in preserving and accessing historical Asian knowledge.