Automatic Image Annotation with Cooperation of Concept-Specific and Universal Visual Vocabularies