Database Reference
In-Depth Information
Fig. 8.17 The re-ranking procedure for modification of a query for video retrieval on a peer-to-
peer network; ( a ) the user node sends query packets to its peers in the neighborhood community;
( b ) the query is modified by the peer nodes and sent back to the user node; ( c ) the user node gathers
the modified queries and computes the second modified query and sends it to the peer nodes; ( d )
the final retrieval results (video files) are sent to the user node
l x i
N
l x i
can be obtained. In this way, the label l j
represents the occurrence of template g j in the input video. Thus, the term-vector
model [ 360 ] can be applied to the resulting set of labels. The number of times g j
present in the video can be viewed as the term frequency (TF) . We can formally
formulate the corresponding term-weighting vector as: v
,...,
1 ,
l j ∈{
1
,
2
,...,
J
},
j ,
j , ʷ
1
i
=
t
where w j represents the multiplication of the term frequency (TF) and the inverse
term frequency (ITF) of the j -th template g j for the input video.
=[
w 1 ,...,
w j ,...,
w J ]
8.5.3
Re-ranking Approach to P2P Video Retrieval
Each node in the P2P network shown in Fig. 8.17 stores a collection of video
clips, each of which is indexed by a weight vector. All the members of the
neighborhood community are assuming the use of a single set of visual templates
C
J . The peers may have a different number of video files.
Figure 8.20 illustrates the zoom version of Peer n . Video indexing within this
peer can be viewed as a network of three layers: (first) the query vector, (second)
= g j |
j
=
1
,
2
,...,
Search WWH ::




Custom Search