MUG FILE PROJECT REPORT NUMBER UHMUG-2 - AN ANALYSIS OF PROCEDURES FOR GENERATING FACIAL IMAGES

• NCJ Number: 47585
• Author(s): K R LAUGHERY; G C DUVAL; R H FOWLER; B T RHODES
• Date Published: Unknown
• Length: 199 pages
• Grant Number(s): 76-NI-99-012
• Sponsoring Agency: US Dept of Justice
• Corporate Author: University of Houston
• Sale Source: National Institute of Justice/
• Language: English
• Country: United States of America

Annotation

THE USE OF SKETCH ARTISTS AND THE IDENTI-KIT AS MEANS OF GENERATING TARGET IMAGES IS STUDIED USING WHITE AND BLACK MALES AND WHITE FEMALES AS THE TARGET POPULATIONS. METHODS EMPLOYED BY WITNESSES AND RESULTS ARE EXAMINED.

Abstract

A TOTAL OF 182 IMAGES WAS GENERATED FOR 97 DIFFERENT PERSONS IN THESE UNIVERSITY OF HOUSTON STUDIES. IN THE STUDY OF BLACK MALES AND THE STUDY OF WHITE FEMALES, 20 TARGET PERSONS AND 40 WITNESSES WERE USED, GENERATING 1 SKETCH AND 1 IDENTI-KIT COMPOSITE FOR EACH TARGET. IN THE STUDY OF WHITE MALES, 57 TARGET PERSONS AND 102 WITNESSES WERE USED. IN ADDITION TO THE IMAGES, A VARIETY OF DATA IS PRESENTED ABOUT THE ATTITUDE OF THE TARGET AND ITS EFFECT ON IMAGE PERCEPTION, THE EFFECT OF VARIOUS WITNESS CHARACTERISTICS, AND THE IMAGE-GENERATING PROCESS. THE STUDY FOUND THAT SKETCHES ARE BETTER REPRESENTATIONS THAN IDENTI-KIT COMPOSITES, THE PROBLEM BEING THAT THE KIT ITSELF LIMITS THE WITNESSES' ABILITY TO MODIFY FACIAL CHARACTERISTICS. CORRELATIONS BETWEEN THE 'GOODNESS-OF-FIT' MEASUREMENTS MADE BY PERSONS AND BY THE COMPUTER ALGORITHM WERE INSIGNIFICANT. THIS INDICATES THAT PEOPLE USE DIFFERENT INFORMATION IN JUDGING SIMILARITY. IT IS SUGGESTED THAT FUTURE ALGORITHM DEVELOPMENT MIGHT ATTEMPT TO INCORPORATE HEURISTICS THAT PARALLEL THE PROCESS USED BY PEOPLE. THE PROCESS ANALYSIS SHOWED THAT PERSONS WORKING WITH ARTISTS SPENT MORE TIME IN THE IDENTITY PROCESS, USED A GREATER NUMBER OF FEATURE CODES, AND MOVED AROUND BETWEEN FEATURES MORE THAN THOSE WORKING WITH AN IDENTI-KIT TECHNICIAN. THE STUDY ALSO CONFIRMED THAT DESCRIPTIONS BY SOMEONE OF THE SAME RACE ARE MORE ACCURATE THAN CROSS-RACIAL DESCRIPTIONS. IMPLICATIONS OF THIS WORK FOR THE COMPUTER-AIDED IDENTIFICATION PROJECT ARE DISCUSSED. DESCRIPTIONS OF THE EXPERIMENTAL METHOD AND THE DATA GENERATED BY THE EXPERIMENTS ARE INCLUDED. APPENDIXES GIVE SAMPLE SKETCHES, THE FORMS USED IN THE EXPERIMENTS, QUALIFICATIONS OF THE ARTISTS AND IDENTI-KIT TECHNICIANS, SAMPLES OF THE COMPUTER ANALYSIS, AND OTHER SUPPORTING INFORMATION. (GLR)