SPEECH-BASED VISUAL CONCEPT LEARNING USING WORDNET (FriAmOR1)

Author(s) : Xiaodan Song (University of Washington, United States of America) Ching-Yung Lin (IBM T. J. Watson Research Center, United States of America) Ming-Ting Sun (University of Washington, United States of America)

Abstract : As the amount of video data increases, organizing and retrieving video data based on their semantics is becoming more and more important. In this paper we approach this problem by developing context-independent keyword lists for interested concepts from WordNet. Furthermore, an approach called common information gain is proposed to reorder and extend the supervised learning generated keyword list by the one generated from WordNet. Experimental results show that the context-independent approach can achieve comparable performance to conventional supervised learning algorithm, and the approach based on the extended keyword list achieves about 53% and 28.4% relative improvement over the best speech-based retrieval algorithm in TRECVID 2003 on dataset CF1 and CF2 respectively.

Menu