Automatic annotation of flash MOVIES with a fuzzy semantic network

In order to address the problem of automatically annotating Flash movies with semantic keywords, we present in this paper FUSE¹ network, which tries to interpret multimedia components semantically with multiple concepts by specifying the degrees of their membership to the concepts individually. The fuzzy semantic network is constructed using both WordNet and scenes extracted from a collection of annotated Flash movies. We also employ relevance feedback to train the network and improve the annotation accuracy progressively. An experimental study is presented based upon a content-based automatic Flash annotating system, which shows that effective Flash annotation performance can be achieved with this approach. Compared with the brute-force manual annotation approach, our work provides a preliminary understanding of the basic content of Flash movies at a lost cost, which is also applicable to other rich media formats, such as SMIL, PowerPoint, and MPEG-7.