In this paper, we discuss semantic image annotation and propose a novel approach, called EMERGSEM, based on emergent image semantics and a recommendation system. The emergent semantics of images are derived from a generic ontology and are generated collaboratively by a group of annotators who assign keywords from a predefined lexical dictionary to images. The resulting instantiated semantic concept graph is used to interpret and relate image objects. In addition, a recommendation system based on a Galois lattice is used to classify user preferences to determine final recommendation lists by finding similarities between correlated groups of user profiles.