Semantic descriptions of non-textual media available on the Web can facilitate retrieval, re-use and presentation of media assets. Semantic Web languages can represent controlled vocabularies and shared annotations of media content on the Web. By identifying concepts to consider, Uniform Resource Identifiers (URIs) are the building blocks of the Semantic Web. Often, particular regions of an image or particular sequences of a video need to be localized and uniquely identified in order to be used as subject or object resource in an RDF annotation. In this talk, we will first present the Media Fragment URI specification, a recent W3C recommendation that enables to uniquely identifying sub-parts of media assets in the same way that the fragment identifier in the URI can refer to part of an HTML or XML document. We will then describe models and ontologies that we will illustrate with several real world applications using semantic annotations attached to media fragments.