A Tool for Content-Based Image Retrieval in Object-Oriented Databases

Conventional image processing algorithms used to compare images are very timeconsuming, making them inappropriate for use in searching for an image in a huge collection of images stored in a database. The usual approach uses icons and textual attributes stored with the images to specify the queries. However, this approach depends on human analysis of each image, which reduces the precision to a great extent. This work states that it is possible to quickly retrieve a set of characteristics from each image as it is being stored in the database, building an index structure based on each characteristic, so images can be retrieved through queries based on the graphic contents of the stored images. When a search for a specific image becomes necessary, these same characteristics can be extracted from that image, and the set of indexes can be used to narrow the search space of the target images progressively. This work describes how these concepts were used to implement a tool that enables the user to issue commands to retrieve images stored in an Object Manager based on their contents. It is aimed at the storage of medical images as one kind of data associated to objects. The objectmodel underlying this tool defines the “image data type,” a novel approach that permits a significant reduction of the set of images to be processed, based on their graphical contents, so the search can be accelerated when queries are issued.