1994 Volume 36 Issue 4 Pages 233-238
This paper presents a computer-based system for retrieving a witnessed face from a database of faces. To construct the database, a male adult was asked to observe 100 male adult faces one by one on a TV monitor and describe the features of each face using a set of 605 descriptors which had been collected in the previous study. Feature descriptors thus chosen for each face are stored in the database. A computer algorithm calculates a similarity value between a description by a witness and the stored description of each face, to search for the faces which match the witness's description. The similarity is defined as a function of inter-feature similarities, using a vector representation of features. These inter-feature similarities are obtained from either human ratings or statistical correlations. An experimental evaluation of the system using 12 college students as “witnesses” gave promising results.