Scale-space multi-view bag of words for scene categorization