Indexing, Object Segmentation, and Event Detection in News and Sports Videos - Multimedia Database Retrieval: Technology and Applications - page 175

Let I Group be the video interval at the group level, which contains a total of S 1
video shots, I Group = { I 1 , Shot ,... I i , Shot ,..., I S 1 , Shot } . Then, the video descriptor for

I Group is given by:

D I Group ≡

t

v Group =[ w 1 , Group ,..., w r , Group ,..., w R , Group ] (7.16)

where

S 1
i = 1 w ir , Shot

=

w r , Group (7.17)

w ir , Shot is the r -th weight component of the i -th video shot I i , Shot . Equations ( 7.16 )
and ( 7.17 ) are also applied to obtain a set of descriptors for a video interval at the
story level, D I Story . A summary of the video description databases for each level is as
follows:

D I i , Shot ,

VD Shot = { (

) | ∈

I Shot (

) }

I i I i F (7.18)

D I i , Group ,

VD Group = { ( I i ) | I i ∈ I Group ( F ) } (7.19)

D I i , Story ,

VD Story = { ( I i ) | I i ∈ I Story ( F ) } (7.20)

In the querying process, a query can be chosen from VD Shot or VD Group ,
according to the links in Fig. 7.3 . In these conditions, a query interval should have
two properties: first, it should be short enough to not contain many lengthy scenes.
Second, it should be long enough such that the context does not break down. This
means that an interval from a story database may not be a suitable interval query.

Fig. 7.3 Multiple-level
access to video database,
( a ) shot-to-shot,
( b ) shot-to-group, ( c ) group
to-group, ( d ) group-to-story,
( e ) shot-to-story