Statistical Examination of Handwriting Characteristics Using Automated Tools

• NCJ Number: 241743
• Author(s): Sargur N. Srihari
• Date Published: February 2013
• Length: 85 pages
• Grant Number(s): 2010-DN-BX-K037
• Agencies: NIJ-Sponsored
• Publication Type: Research (Applied/Empirical)

Annotation

This report presents the results of research aimed at developing new statistical methods and software tools for examining handwriting characteristics.

Abstract

This report presents the results of research aimed at developing new statistical methods and software tools for examining handwriting characteristics. The research consisted of six goals: 1) develop methods to extract samples of commonly encountered letter forms from extended handwriting samples of typical writers in the United States, 2) prepare the appropriate format to present the samples to QD (questioned document) examiners who would then enter perceived characteristics with a user interface, 3i) determine the frequency of occurrence of combinations of handwriting characteristics, 4) use those frequencies to construct a probabilistic model without the method being overwhelmed by the combinatorial possibilities and sample requirements, 5) develop methods to infer the probability of evidence from the model, and 6) indicate where such methods could be used in the QD examiner's work-flow. The project's tasks were divided into four parts, and are presented in that format in this report: data preparation, model construction, inference, and QD work-flow. Several conclusions were reached as a result of this project. These conclusions include 1) statistical characterization of handwriting characteristics can be useful to assist the QD examiner in the examination of handwritten items; 2) since probability distributions of handwriting characteristics involve too many parameters, the complexity can be handled using probabilistic graphical models (PGMs); 3) PGMs can be used to determine the rarity of given characteristics; 4) software interfaces for creating databases of handwriting characteristics for different commonly occurring letter combinations have been developed, 5) an automatic method for determining handwriting type was introduced, with a 92 percent accuracy, and 6) statistical methods can be used in the work-flow of the forensic document examiner. Implications for policy and practice are discussed. Tables, figures, algorithms, appendixes, and references