In this paper we present a novel infrastructure for querying heterogeneous databases composed of semantically annotated documents, using XML technology. We have developed a query generalization system in order to make the use of different XML schemas transparent. It uses a preliminary specification of elements considered equivalent in the various schemas. By defining different levels of equivalence, the system is able to translate queries between different annotated XML schemas. We have tested the system on the particular case of image database, more precisely, we consider different tagged image formats. We obtain query translation that preserves the semantics and offers a good match.