Clustering writing styles with a self-organizing map

Abstract

This work shows how a self-organizing map (SOM) can be applied in the analysis of different handwriting styles. The handwriting samples analyzed have been collected in online fashion with special writing equipments such as pressure sensitive tablets. The handwriting style of an individual subject is represented by a vector components of which reflect the tendencies of the writer to use certain prototypical styles for isolated alphanumeric characters. This study shows that correlations between different writing styles, both character-wise and writer-wise can be found. Clusters of different personal writing styles can be found by studying the U-matrix visualization of the SOM trained with data collected from over 700 subjects. An examination of the component planes of the SOM reveals some interesting correlations between the prototypical character styles.