Formalization and management of content-based information retrieval of complex objects with hierarchical and cross-modal data

Abstract

Image based reports, such as papers, web pages and thesis have been digitally stored, with fast growing in volume and complexity. This scenario challenges both database management systems' efficiency and effectiveness. Big data addresses issues found when dealing with great data volume, scope, complexity and distribution. On the other hand, as complexity grows, new query methods are needed - numeric data and short strings are not enough to represent complex data. Similarity search has been proven to be the state-of-the-art to compare complex data. However, the concept of similarity itself becomes unclear when objects under comparison are composed of parts that can be compared by similarity as well, i.e., composite objects. There are several studies covering the similarity search over a single object, as images, audio and genetic sequence. However, currently there are no methods on the literature that deal with the similarity of composite objects. Moreover, similarity is defined by considering the semantic meaning to treat few data. This project aims to develop techniques to compare composite and complex objects that can be hierarchically organized regarding their similarity. The proposed techniques will be evaluated in different domains, as medical exams with several images; digital library, aiding recommender systems of academic thesis. (AU)