Title:   "Context-based news video retrieval"

Speaker: Ichiro IDE (Nagoya Univ / National Institute of Informatics, Japan)

Abstract:
Due to progress in various fields of information technology, a mass amount 
of multimdia data have become available to us. However, current technology
still does not provide sufficient tools to retrieve what a user wants 
effectively and efficiently. My approach to this problem is to annotate 
videos with text-based indices (keyphrases) that reflect the visual 
contents of the image, and by providing an interface that filters out 
redundant information. The work focuses especially on news video since it 
is a continuous record of the human activity, which could be considered as
a precious historical heritage.
I will introduce three works I have done in the past:
(1) Attribute-based news video indexing; a context-based indexing method 
that checks the coherency of the attributes derived from image features 
and keyphrase candidates obtained from text. 
(2) Topic threading; a method that extracts topic structures within a news
video archive, which organizes the structure so that according to time and
topic relations. The thread structure is used in the backend of a topic 
retrieval interface, so that a user can follow down topics as if following
a thread of his/her interest.
(3) trackThem interface; a news video browsing interface based on human 
relationships extracted from a news video archive.