An interactive image management system for face information retrieval

Abstract

The complex nature of two-dimensional image data has presented problems for traditional information systems designed strictly fo alpha-numeric data. In many systems, images are simply treated as an additional data type, making no use of the actual image data. However, this image data provides an enormous amount of information and should be utilized for more powerful retrieval capabilities. In this paper we present a general system architecture for image management systems which combines the strengths of computer vision systems with traditional information management techniques. This system will provide a general framework for the development of any domain-specific application. Our approach utilizes computer vision routines for both insertion and retrieval, and allows easy query-by-example specifications. The vision routines are used to segment and evaluate images based on descriptions of the domain objects and their attributes. The vision system can then assign feature values to be used for similarity-measures and image retrieval. An image management system developed for face-image retrieval is presented to demonstrate these ideas.