Information Technology Reference
In-Depth Information
Fig. 11.8 Screenshot of Online Analysis view of our visualization tool (showing classification
results)
(e.g., containing a man giving a speech or featuring strong music in the background)
are also easily classified as nonviolent. On the other hand, the method wrongly
classifies a video segment as violent when the segment contains very strong sounds
or exciting moments such as a plane taking off or a bell ringing loudly. The most
challenging violent segments to detect are the ones which are “violent” according
to the objective definition of violence given in the MediaEval VSD task, but which
actually contain only actions such as self-injuries, or other moderate actions such as
an actor pushing or hitting slightly another actor. Our method is also unable to detect
violent video segments which are “violent” according to the objective definition
of violence, but which contain no audio cues exploitable for the identification of
violence (e.g., a man bleeding). More detailed discussion on the performance of our
method is given in [ 1 ].
11.6 Conclusions and Future Work
In this chapter, we presented an approach for the detection of violent content in
movies and short web videos at the video segment level. We employed low and
mid-level audio-visual features to represent videos. The mid-level audio and visual
Search WWH ::




Custom Search