Distributed ContentBased Visual Information Retrieval System on PeertoPeer Networks

1
Distributed Content-Based Visual Information
Retrieval System on Peer-to-Peer Networks
2
Outline

• Introduction
• Firework Query Model
• Experiment and Result
• Conclusion

3
Introduction

• Property of P2P network
• There are no dedicated servers and clients.
• Every peer may act as server and client depending
  on situation.
• P2P networks are not centrally managed.

4
Introduction

• Advantage of P2P network for CBIR
• More image collection
• Distributed computing

5
Introduction

• Information retrieval in a P2P network

6
Peer Clustering

• it does not have a centralized coordinator to
  maintain the feature vector index.
• It must perform a brute force search,which is
  inefficient.
• It propose to cluster peers with similar image
  features together.

7
Peer Clustering

• Definition 1Let Collection(p) be
  the set of n images a peer p shares.
• where f means a specific feature extraction
  function.
• After the extraction,each peer contains a set of
  feature vectors
• where
• d is the number of the dimension.

8
Peer Clustering

• Definition 2Cat(p) is defined as
  where and are the mean and variance of the
  image feature vectors collection .
• The j-th mean and variance is defined as

9
Peer Clustering

• Definition 3 Sim(p, q) is defined as the
  distance measure between two peers signature
  values.
• The following formula is used
• Two peers p and q are similar,when Sim(p, q) is
  small.

10
Peer Clustering
11
Firework Query Model Over Clustered Network

• Query massage reaches its designated cluster, the
  query message is broadcast by peers through the
  attractive connections inside the cluster.

12
Shared File Look Up

• Let be the query feature vector. is
  the distance measure between the query and an
  image i shared by peer p.
• It is an L2 norm defined as,

13
Route Selection

• The peer calculates the similarity between the
  query and its signature value, which is
  represented as,

14
Routing Algorithm

• Firework-query-routing (peer p, query Q)
• if Sim(p, Q) gt (threshold) then
• reply the query Q
• if rand() gt CTS then
• TTL new(Q) TTLold(Q) -1
• end if
• forward Q to all linkattractive(p)
• else
• TTL new(Q) DTTL old(Q) - 1
• if TTL new(Q) gt 0 then
• forward Q to all linkrandom(p)
• end if
• end if

15
Routing Algorithm

• TTLTime -To-Live
• For attractive connections, the probability of
  decreasing the TTL value is an arbitrary value in
  0, 1 called Chance-To-Survive (CTS).
• This strategy can reduce messages passing outside
  the target cluster,so more relevant information
  can be retrieved.

16
Experiment and Result

• Performance Metrics
• Recall RecallRa/R ,
• where Ra is the number of retrieved relevant
  documents,and R is the total number of relevant
  documents in the P2P network.

17
Experiment and Result

• Performance Metrics
• Query scope VisitedVpeer/Tpeer ,
• where Vpeer is the number of peers that
  received and handled the query and Tpeer is the
  total number of peers in the P2P network.
• Query efficiency EfficiencyRecall/Visited
• Recall and Query Scope are not enough to
  determine the quality of the algorithm.

18
Experiment and Result

• Recall versus number of peers

19
Experiment and Result

• Query scope versus number of peers Query
  efficiency versus number of peers

20
Experiment and Result

• Recall versus TTL value of query
  Query scope versus TTL value of query

21
Experiment and Result

• Query efficiency against TTL value of query
  packet

22
Conclusion

• The architecture fully utilizes the storage and
  computation capability of computers in the
  Internet.
• the lack of a centralized index requires a query
  to be broadcast throughout the network.
• we propose a peer clustering and intelligent
  query routing strategy to search images
  efficiently over the P2P network.