Uso de ontologias como apoio ao processo de recuperação de imagem por conteúdo baseado em descritores

Abstract

Facts such as the spread of personal computers and Internet,allied with easy-access and lowcost digital image capture devices, have lead to an exponential growth on image databases. Along this huge source of visual information, problems related to the queries of this content arose. The owners of the image repositories or clients of the content’s tools became to demand faster solutions when querying for some certain visual patterns. With the purpose of fulfilling this increasing demand , the CBIR (Content Based Image Retrieval) was esigned. The latter consists on the extraction of color, shape and texture features from images and the application of mathematical relations which enables the comparison between images, aiming to respond to queries. However, this methodology faces semantic problems: The representation of the image features and its comparison is not intuitive to the user, which disturbs the relationship between the query results and and their proper outcome . Aiming to handle this problem, we propose the use of ontologies to support the descriptor-based retrieval process, in order to make the query process more intuitive. In addition, we introduce an experiment-driven software , which might be expanded to another approaches. The chosen image dataset is comprised of flower species and has been used to analyze the descriptor-based query process and to propose how an ontology might improve it.