HQA²LFS-handwriting quality assessment using an active learning framework in smartphones.

Area of Science:

• Computer Science
• Forensic Science

Background:

• Assessing handwriting quality is crucial for various applications, including forensic analysis and educational evaluation.
• Existing methods often lack objective and quantitative measures for word-level handwriting assessment.

Purpose of the Study:

• To develop and evaluate a regression-based approach for assessing handwriting quality at the word level.
• To identify key features influencing handwriting quality and compare different machine learning models.

Main Methods:

• Utilized a diverse corpus of over 1296 unruled and 1160 ruled handwriting samples from 65+ writers.
• Employed structural, perceptual, and fringe features for analysis.
• Trained and compared various machine learning models, including Random Forest and XGBoost, incorporating an active learning strategy.

Main Results:

• Random Forest and XGBoost models achieved an excellent R-squared value of 0.996.
• Perceptual attributes (neatness, readability) were the most influential features.
• Active learning enhanced model training by effectively selecting uncertain samples compared to random sampling.

Conclusions:

• The developed regression-based approach accurately assesses word-level handwriting quality.
• The findings highlight the importance of perceptual features and the efficacy of active learning in model refinement.
• The method provides a robust tool for analyzing handwriting characteristics and identifying low-quality manuscripts.