SQL-Based Compound Object Comparators: A Case Study of Images Stored in ICE

We introduce the framework for storing and comparing compound objects. The implemented system is based on the RDBMS model, which - unlike other approaches in this area - enables to access the most detailed data about considered objects. It also contains ROLAP cubes designed for specific object classes and appropriately abstracted modules that compute object similarities, referred as comparators. In this paper, we focus on the case study related to images. We show specific examples of fuzzy logic comparators, together with their corresponding SQL statements executed at the level of pixels. We examine several open source database engines by means of their capabilities of storing and querying large amounts of such represented image data. We conclude that the performance of some of them is comparable to standard techniques of image storage and processing, with far better flexibility in defining new similarity criteria and analyzing larger image collections.