|
Activity Number:
|
294
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, July 31, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
IMS
|
| Abstract - #309241 |
|
Title:
|
Methods for Constructing Classifiers from Sparse Categorical Data for Forensic Document Analysis
|
|
Author(s):
|
Christopher Saunders*+ and Linda J. Davis
|
|
Companies:
|
George Mason University and George Mason University
|
|
Address:
|
4400 University Drive Ms1G8, Fairfax, VA, 22030,
|
|
Keywords:
|
Categorical Data ; Forensic Document Analysis ; Pattern Recognition ; Bayes classifiers ; Biometric identification
|
|
Abstract:
|
Recent challenges to the admissibility of handwriting evidence have illustrated the need for a statistical approach to the problem of writer identification. Discriminant functions are constructed with a dataset containing one hundred distinct writers who have each submitted five script copies of the London paragraph. Each character in the paragraph is represented as a graphical isomorphism; reducing each paragraph to the frequency of graphs used to write each character. This presentation compares approaches to constructing plug-in Bayes classifiers based on the categorical data that predict the writer-ship of an unknown document. The classifiers are compared using leave-one-out cross-validation.
|
