MEDIAEVAL 2015, Multimedia Benchmark Workshop, September 14-15, 2015, Wurzen, Germany
This paper describes our approach to carry out multimedia search by connecting the textual information, or the corresponding textual description of the required visual content, in the user query to the audio-visual content of the videos within the collection. The experiments were carried out on the dataset of the Search and Anchoring in Video Archives (SAVA) task at MediaEval 2015, consisting of roughly 2700 hours of the BBC TV broadcast material. We combined visual concepts extraction con dence scores with the infor-mation about corresponding word vectors distances in order to rerank the baseline text based search. The reranked runs did not outperform the baseline, however they exposed potential of our method for further improvement.
Type:
Conférence
City:
Wurzen
Date:
2015-09-14
Department:
Data Science
Eurecom Ref:
4688
Copyright:
CEUR
See also:
