Video retrieval by using video ontology

点击翻译按钮获取中文摘要

In videos, the same event can be taken by different camera techniques and in different situations. Therefore, shots of the same event may contain significantly different features. In order to retrieve such diverse sets of shots for a given event (query), we propose a method which defines an event based on the rough set theory. First, given subsets of shots for an event as positive examples, we represent the event as the union of the subsets. Then, we adopt a partially supervised learning approach to obtain negative examples from a large amount of unlabeled data. To be precise, we identify "likely" negative examples from the unlabeled data based on their dissimilarities to the given positive examples. In calculating dissimilarities, we take advantage of subspace clustering to find clusters in different subspaces of the high-dimensional feature space.