Bagging-based cross-media retrieval algorithm

It is hard to come up with a strong learning algorithm high cross-media retrieval accuracy, but finding weak slightly higher accuracy than random prediction simple. This paper proposes an innovative Bagging-based (called BCMR) based on this concept. First, we use bootstrap sampling take sample from the original set. The amount of abstracted by bootstrapping set be same as dataset. Secondly, 50 replicates are used train classifiers. In our experiments, homogeneous individual classifiers and eight different baseline methods. Last not least, generate final classifier using voting integration strategy. Using collective wisdom, can get rid bad decisions, giving integrated model much better generalization ability. Extensive experiments three datasets show that BCMR significantly improve accuracy.