Index-based object recognition in pictorial data management

Modern database management systems are designed to promote the controlled sharing of information. Each individual user has his own view of the data and acts as if he were the only user of the system. While database technology has emphasized the management of textural data, in the past few years there has been much interest in the ability to manage non-textual data. This interest from the database community has manifested itself in proposals for various types of multimedia databases, while from the computer vision community it has manifested itself in the design of image databases [Gro84, Na85, TaY841. Our philosophy behind the design of such databases has differed from those of other researchers in this area. The computer vision community has designed image database management systems either as testbeds for their algorithm development or for geographical applications, whereas the database community has, by and large, treated images as just another field of a record. We believe on the other hand, that images and text should be treated equally. The backbone of our approach is image interpretation. If images should be able to be components of queries, it seems clear to us that image interpretation should be a component of a viable query processing strategy. Both research communities have something to offer to this field. The database community is undergoing a conceptual revolution which has lately been overtaking the relational paradigm. This new paradigm concerns itself with managing objects [BanSS, BCG86, CDF86, Dit86, Row87]. An object is an instantiation of an abstract data type which is embedded in a multiple inheritance network. As such, an object is packaged with some predetermined operations. Instead of managing