Text-independent writer identification and verification using textural and allographic features

Area of Science:

• Behavioral biometrics
• Forensic science
• Document analysis

Background:

• Handwriting identification is a key biometric modality.
• Applications include forensic and historical document analysis.
• Behavioral biometrics research is crucial for unique human identification.

Purpose of the Study:

• To develop novel techniques for automatic writer identification and verification.
• To create methods independent of the textual content of handwriting samples.
• To enhance the accuracy of writer identification using advanced feature extraction.

Main Methods:

• Utilizing probability distribution functions (PDFs) extracted from handwriting images.
• Analyzing handwriting at texture level using contour-based joint directional PDFs.
• Analyzing handwriting at character-shape (allograph) level using grapheme PDFs.

Main Results:

• Developed effective techniques for writer identification and verification.
• Achieved writer characterization independent of textual content.
• Combined directional, grapheme, and run-length PDFs for improved performance.

Conclusions:

• The proposed methods offer practical feasibility for free-style handwriting analysis.
• Reliable probability estimates can be obtained from a few lines of handwritten material.
• The techniques provide an intimate characterization of individual handwriting style.