DeepJiandu Dataset for Character Detection and Recognition on Jiandu Manuscript

Area of Science:

• Digital Humanities
• Historical Linguistics
• Computer Vision

Background:

• Ancient Chinese Jiandu (bamboo and wooden slips) are crucial historical records predating paper.
• Accurate Jiandu character recognition is vital for historical research and manuscript processing.
• Existing methods face challenges due to character degradation and varied forms on Jiandu.

Purpose of the Study:

• Introduce the DeepJiandu dataset for Jiandu character detection and recognition.
• Provide a reliable resource for training and evaluating Jiandu character recognition models.
• Facilitate the study and organization of historical Jiandu manuscripts.

Main Methods:

• Developed a dataset of 7,416 images.
• Annotated 99,852 characters across 2,242 categories.
• Dataset designed to address challenges like degradation and variable character forms.

Main Results:

• The DeepJiandu dataset offers authentic and reliable data.
• It effectively handles complex Jiandu character recognition challenges.
• Enables robust model training and evaluation for Jiandu analysis.

Conclusions:

• The DeepJiandu dataset is a significant resource for Jiandu research.
• It will accelerate the deciphering and organization of ancient Chinese historical records.
• Improves the accessibility and understanding of Jiandu cultural heritage.