Feasibility Study of Visualizing Diversity of Japanese Hiragana Handwritings by Multidimensional Scaling of Earth Mover's Distance toward Assisting Forensic Experts in Writer Verification

Abstract

In this paper, we demonstrated a mapping procedure to visualize the diversity of overall handwriting shapes of five Japanese Hiragana characters for the purpose of assisting forensic examiners in the process of writer verification. Multidimensional scaling was applied to Earth Mover's Distance (EMD) data calculated between 60 different writers in order to visualize each writer's feature in population. EMD flow was calculated between k-means cluster centroids, which are representative points of kernel density distribution of handwritten stroke of each writer within six trials. Experimental results showed that the relative relation of overall handwritten shapes of each writer was successfully visualized as the locus in multidimensional space. The state of distribution such as the density in multidimensional space is considered to provide effective information to forensic examiners in evaluating the rarity of handwritten features observed in questioned document.