Efficient semantic annotation method for indexing large personal video database

As there is a large gap between high-level semantics and low-level features, it is difficult to automatically obtain high-accuracy video semantic annotation through general statistical learning based methods. In this paper, we propose a novel annotation framework based on active learning and semi-supervised ensemble method, which is specially designed for personal video database. To efficiently annotate the home video database, an initial training set is first elaborately constructed based on the distribution analysis of the entire video dataset. Then, both a semi-supervised ensemble based method and an active learning based method are proposed, which aims at minimizing a margin cost function of ensemble to ensure the generalization capacity. The experiment results on about 50 hours home videos show that the proposed method performs superior to both existing semi-supervised learning algorithms and the general active learning algorithms in terms of annotation accuracy and performance stability.