Abstract
Content-Based Image Retrieval Systems (CBIR) aim at helping in searching large image collections to find those more likely to answer query conditions based on the information represented in the images. To speed up the search process, selected features are extracted from each image when they are stored in the database, so each one is represented by a feature vector. Subsequent image searching operations are performed using the feature vectors in place of the images. The feature extraction algorithms have important issues in CBIR due to the large semantic gap between the low-level features extracted as compared to the high-level, semantic, results expected by the users. A way to approach a semantic analysis of an image, as performed by humans, is to employ a large number of analyzers whose results are processed by a set of rules based on if-then clauses. In this paper we create a framework to define image processing and feature extraction algorithms for CBIR systems as components of a data ow architecture, including an analysis mechanism to interpret images. It is based on decision components that check the results of previously executed feature extractors to choose from a set of configurable execution paths that leads to the creation of the feature vector of each image. In the paper we will describe a real system that has been implemented based on these concepts, which is being used as a teaching tool in a school hospital.