NCJ Number
57754
Date Published
1977
Length
4 pages
Annotation
THIS REPORT DESCRIBES A METHOD OF CLASSIFICATION OF PICTURES OF HUMAN FACES FOR THE PURPOSES OF QUICK IDENTIFICATION OF A GIVEN PICTURE USING COMPUTERS.
Abstract
THE MATHEMATICAL METHOD USED IS BASED ON A VERSION OF FACTOR ANALYSIS CALLED 'OPTIONAL FACTOR ANALYSIS.' A MATHEMATICAL MODEL WAS DEVELOPED AND TRANSLATED INTO A COMPUTER PROGRAM BASED ON COMPARISONS OF CERTAIN PROPORTIONS OF THE HUMAN FACE AND ADDITIONAL INFORMATION ABOUT THE FACE FROM EYEWITNESSES. IN ALL, TWO COMPUTER PROGRAMS WERE PREPARED. THE FIRST PROGRAM, WHICH DEALS WITH THE CLASSIFICATION PROCESS, IN A CLUSTERING ALGORITHM BASED ON THE FACTOR ANALYSIS METHODS. THE SECOND PROGRAM DEALS WITH THE RECOGNITION PROCESS ITSELF; WHEN A PICTURE IS DESCRIBED TO THE COMPUTER, THIS ALGORITHM RANKS THE PICTURES IN THE ALBUM ACCORDING TO A DEGREE OF SIMILARITY WITH THE PICTURE TO BE IDENTIFIED. THE COMPUTER PROGRAM PACKAGE WAS TESTED ON A PILOT ALBUM OF ABOUT 500 PICTURES. THE ALBUM WAS ENCODED AND FED INTO THE COMPUTER AND THEN THE ALBUM'S 506 MUGSHOTS WERE COMPARED WITH 12 IDENTI-KIT IMAGES. IN ALL 12 CASES, THE IDENTITY OF THE CRIMINAL REPRESENTED BY THE IDENTI-KIT IMAGE HAD ALREADY BEEN DISCOVERED BY CONVENTIONAL MEANS, AND THE PHOTOGRAPHS OF THE ACTUAL CRIMINALS WERE RANDOMLY SCATTERED THROUGHOUT THE PILOT ALBUM. THE COMPUTER WAS ASKED TO IDENTIFY EACH OF THE 12 UNKNOWN FACES, WHICH WERE ENCODED ACCORDING TO THE FACIAL MEASUREMENTS AND THE RELEVANT VERBAL DESCRIPTION. OF THE 12, 3 WERE CORRECTLY IDENTIFIED WITH THE HIGHEST SIMILARITY COEFFICENT, WHILE 2 OTHERS PLACED 2ND, 1 PLACED THIRD, 2 PLACED 4TH, AND ANOTHER AT THE 6TH HIGHEST COEFFICIENT. REFERENCES, EQUATIONS, AND OTHER ILLUSTRATIONS ARE INCLUDED. (KBL)