Quantifying Writer Variance Through Rainbow Triangle Feature Extraction

Abstract

Handwriting analysis is conducted by forensic document examiners who can visually recognize specific landmark writing characteristics to determine writership of a document. Within the past ten years, there has been strong criticism of handwriting analysis as a practice, which has motivated researchers to develop methods to quantify similarities between two written documents. This would better support conclu- sions drawn by current forensic document examiners. This project explores this possibility through a combination of a software package written in R called handwriter and a novel analysis method which we call Rainbow Triangle Feature Extraction to accomplish these goals. We hypothesize that given a sufficient number of handwritten docu- ments from a sample of writers, we can calculate a score-based likeli- hood ratio for the common source versus different source hypotheses. The first step in the analysis is to use handwriter to decompose a scanned handwritten document into individual words. Next, we ex- tract all occurrences of the common word ‘the’ and begin the triangu- lation process. To do so, we follow a set of rules to draw segments and triangles that capture the geometry of the word. From these segments and triangles we extract measurements from specific landmarks of the writing which we then use for statistical analysis and comparison.