Efficient image recovery using visual and semantic contents

Images recovery by combining low level descriptors and semantic content is a big challenge. In this work we present a three level software architecture implemented in an Object-Relational database that allows the image queries using both descriptions. For the physical level (low-level) MPEG-7 standard descriptors are employed and for the semantic metadata RDF triplets are used. The tests performed using combined low-level and semantic queries gave very good results in terms of images recovered. Since the semantic data can involve a huge number of RDF triplets, we are proposed and indexation structure for the triplets which shows a very good performance compared with other approaches.