Advancements and Challenges in Handwritten Text Recognition: A Comprehensive Survey

Area of Science:

• Computer Science
• Digital Humanities
• Archival Science

Background:

• Handwritten Text Recognition (HTR) is crucial for digitizing historical documents.
• French civil registers present unique challenges like varied handwriting and annotations.
• Existing HTR systems require evaluation for specialized historical datasets.

Purpose of the Study:

• To survey modern and historical HTR systems for French handwritten text.
• To identify challenges and propose future research directions for transcribing French historical documents.
• To provide an updated overview of HTR research and benchmark datasets.

Main Methods:

• Conducted a literature review of HTR systems across international and French datasets.
• Classified HTR systems by techniques, datasets, publication year, and recognition level.
• Analyzed system accuracies and showcased commercial HTR system performance.
• Summarized publicly available HTR datasets, including ICDAR and ICFHR benchmarks.

Main Results:

• Identified key challenges in transcribing French historical documents, including Belfort civil registers.
• Presented a comparative analysis of HTR system performance, highlighting top approaches.
• Showcased the capabilities of current commercial HTR solutions.
• Provided a comprehensive list of relevant HTR datasets for benchmarking.

Conclusions:

• The survey offers a clear overview of the state-of-the-art in HTR for French historical documents.
• Identified research gaps and potential future directions for improving transcription accuracy.
• Emphasizes the need for specialized HTR approaches for unique archival materials.