based on the existing RDF Dump Format model it isn't possible to match the Media-Info RDF-Dump based on a Wikidata-SPARQL-Query for images using ?item wdt:P18 ?img. The reason is, that the commons media file value stored in the Wikidata Wikibase like http://commons.wikimedia.org/wiki/Special:FilePath/%FileName% but in the media-info rdf-dump this values isn't represented, only the full qualified upload url, which is hard to guess without any other information.