Pseudo-Relevance Feedback for Multimedia Retrieval

摘要: Video information retrieval requires a system to find rele-vant query which may be presented simultaneously in different ways through text description, audio, still images and/or video sequences. The actual search also takes place the text, image or domain. We present an approach that uses pseudo-relevance feedback from retrieved items are NOT similar relevant items. An evaluation on 2002 TREC Track queries shows this technique can substantially improve and performance real collection. believe negative great promise for very difficult multimedia tasks involve similarity, especially when combined with other agents.