VideoSense at TRECVID 2011 : Semantic indexing from light similarity functions-based domain adaptation with stacking