This paper describes a system for content-based retrieval of facial im
ages from an image database. The system includes feature extraction ba
sed on expert-assisted feature selection, spatial feature measurement,
feature and shape representation, feature information compression and
organization, search procedures, and pattern-matching techniques. The
system uses novel data structures to represent the extracted informat
ion. These structures include attributed graphs for representing local
features and their relationships, n-tuple of mixed mode data, and hig
hly compressed feature codes. For the retrieval phase, a knowledge-dir
ected search technique that uses a hypothesis refinement approach extr
acts specific features for candidate identification and retrieval. The
overall system, the components, and the methodology are described. Th
e system has been implemented on an IBM Personal System/2(R) running O
perating System/2(R). Examples demonstrating the performance of the sy
stem are included.