The following query tries to find the categories of an image in a P​18 statement:

# trying to use mwapi to get (license) categories of commons images
# see https://www.mediawiki.org/wiki/Wikidata_query_service/User_Manual/MWAPI
SELECT * WHERE {
  hint:Query hint:optimizer "None".
  BIND(wd:Q668563 as ?leinestr) # random item with picture
  ?leinestr wdt:P18 ?picture.

  BIND(STRAFTER(str(?picture), "Special:FilePath/") AS ?filename)
  BIND(CONCAT("File:", ?filename) AS ?file)

  SERVICE wikibase:mwapi {
    bd:serviceParam wikibase:api "Categories".
    bd:serviceParam wikibase:endpoint "commons.wikipedia.org".
    bd:serviceParam mwapi:titles ?file.
    ?cat wikibase:apiOutput mwapi:category.
  }
}

(See also T168876 for what happens if the optimizer hint is removed.)

However, it finds no results, because the title from the statement is URL encoded (?file is File:U-Bahn%20Berlin%20Leinestra%C3%9Fe.JPG), but the MediaWiki API expects an unencoded title (File:U-Bahn Berlin Leinestraße.JPG). This particular example can be made to work with some REPLACE()s, but clearly that approach doesn’t scale.

One solution would be for the MWAPI to automatically decode titles in parameters. This could happen, for instance, if the title is a full URI as saved in a Wikidata statement (http://commons.wikimedia.org/wiki/Special:FilePath/…). (We probably don’t want to do this for all values in parameters unconditionally, since they might already be decoded but contain % characters.)

Another solution would be to provide a function for URL-decoding (and probably, for symmetry, another one for URL-encoding). This would be more generally useful, but also potentially harder to discover as a solution.