Visual Resampling for Pseudo-Relevance Feedback during Speech-based Video Retrieval.
Stevan Rudinac, Martha A. Larson, Alan Hanjalić · 2009
A method is proposed that makes use of visual reranking to selectively sample feedback sets for Pseudo-Relevance-Feedback during speech-transcript-based video retrieval. Observed per-formance improvement is indicative of the ability of visual reranking to increase the relevance den-sity of the feedback set. 1