A System for Automated Extraction of Metadata from Scanned Documents using Layout Recognition and String Pattern...

Area of Science:

• Digital archiving
• Information retrieval
• Machine learning

Background:

• Manual metadata acquisition is costly and time-consuming for digital document archiving.
• Metadata embedded within documents is crucial for discovery and access.
• Automated methods are needed to efficiently extract this metadata.

Purpose of the Study:

• To develop and describe an automated metadata extraction (AME) system.
• To reduce the cost and improve the efficiency of archiving digital documents.
• To present a novel approach for extracting embedded metadata from textual documents.

Main Methods:

• Developed an AME system combining layout classification/recognition models with a metadata pattern search model.
• Utilized Support Vector Machine and Hidden Markov Models for layout recognition.
• Employed a rule-based metadata search model analyzing string patterns for extraction.

Main Results:

• Successfully extracted metadata from a historic Food and Drug Administration collection.
• Demonstrated the system's effectiveness for structured and semi-structured text corpora.
• The AME system offers a cost-effective solution for metadata acquisition.

Conclusions:

• The AME system provides an efficient and automated approach to metadata extraction.
• The system shows promise for adapting to similar historical document collections.
• Ongoing enhancements aim to further improve the AME system's capabilities.