Computer-assisted handwriting analysis tools being developed for forensic applications
BUFFALO, N.Y. -- Ransom notes, like the one left behind in the JonBenet Ramsey murder case, and other handwritten documents that provide clues to criminal cases may soon be easier to analyze, thanks to research being conducted by University at Buffalo computer scientists.
Researchers in UB's Center of Excellence for Document Analysis and Recognition (CEDAR) have been awarded a $428,000, 16-month grant from the National Institute of Justice to develop computer-assisted handwriting-analysis tools for forensic applications.
The new tools will for the first time make available to law-enforcement investigators quantitative methods for analyzing handwriting in an effort to identify writers of specific documents -- who also may be suspects in criminal cases.
"Our first focus in this project will be to establish on a scientific basis whether or not handwriting is truly individual," said Sargur Srihari, Ph.D., SUNY Distinguished Professor in the UB Department of Computer Science and Engineering and director of CEDAR. "We will be asking 'is the handwriting of different individuals truly distinct?'"
CEDAR is the largest research center in the world devoted to developing new technologies that can recognize and read handwriting. In the United States, it is the only center in a university where researchers in artificial intelligence are applying pattern-recognition techniques to the problem of reading handwriting.
Over the past more than 10 years, CEDAR developed and refined software now used by the U.S. Postal Service to read and interpret up to 80 percent of all handwritten addresses on envelopes. CEDAR researchers today continue to refine and improve the software for the USPS, as well as for Australia Post, which also has adopted the CEDAR system.
That expertise attracted the attention of the National Institute of Justice (NIJ), which has been directed to establish a scientific basis for handwriting testimony in court cases.
The NIJ project requires researchers to look at handwriting from a different perspective than that required by the postal service project.
"Previously we never were interested in who the author was," said Srihari. "The main focus of our Handwritten Address Interpretation system was always to say, 'What is common or average about this handwritten address,' not 'What is special.' But with this project, we will be asking, 'What is special about this?'"
Efforts to analyze handwriting in criminal or civil cases have involved obtaining samples of writing from potential suspects or witnesses and then comparing them with the writing in the document in question.
In the JonBenet Ramsey case, for example, potential suspects, including friends of the Ramseys, were made to write some of the same words that appeared in the ransom note so that investigators could compare them with the original document.
Many less-sensational cases involve forged wills and other handwritten documents.
A 1993 Supreme Court decision stated that in order for expert testimony to be admitted in court cases at any level, a scientific basis for that expertise must be proven through research and the peer-review process.
But because relatively few, if any, objective criteria exist for analyzing handwriting, it has yet to be regarded with the same confidence level as other kinds of evidence.
For instance, human analysts now need to make elaborate measurements about such details of a person's writing as how often a certain slant or loop occurs. The software under development at UB will be able to estimate automatically the slant angle of someone's script, as well as other features of an individual's writing. Those features and quantitative data about them -- such as how often they occur and in what context -- can then be compared with the writing in the document under investigation.
"What we hope to do is to create an automated system that could pick out handwriting styles and attach to them some numbers and confidence levels for a specific document, to evaluate how good a match there is between a sample and a given document," said Srihari.
Srihari and his colleagues at CEDAR are analyzing handwritten addresses gathered from their postal research; they also will be collecting individual handwriting samples from a cross-section of the general population.