CBIR with RF

A Review on Content Based Image Retrieval with
Relevance Feedback using Soft Computing
techniques
PRESENTED BY
RAJ KUMAR JAIN
DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING AND
INFORMATION TECHNOLOGY
MADHAV INSTITUTE OF TECHNOLOGY & SCIENCE,
GWALIOR (M.P.) – 474005

Outline
• Introduction of CBIR
• Problem Statements
• Work in relevance feedback
• Soft computing techniques
• Performance measures
• Conclusion

Introduction
• Image retrieval is a kind of technique of
information retrieval from the huge amount of
database in form of image.
• There are various kind of approach of IR.
1. Text based approach : this approach is used to
annotates the meaning of text into image
form.
2. Content based approach: we provide the
image as a query and extract the feature and
match its similarity from Data base image.

Content based approach of image
retrieval
• There are various kind of approach of feature
extraction.
1. Color feature extraction: it is used the color
feature from an image like RGB color
component , color histogram etc.
2. Texture feature extraction: this feature extract by
several method like gray level co-occurrence
matrix, Tamura feature etc.
3. Shape feature extraction: this feature extract the
edges from an image like edge detection
algorithm, contour detection etc.

Relevance feedback
• RF is an approach to refine the query.
• It reduces the semantic gap between low level
feature and high feature.
• RF is helpful to achieve the better result
according to user perception.

Problem Statement
• The major difficulty of CBIR lies in the big
gap between low-level image features and
high-level image semantics.

• The retrieval efficiency and timing performance of
content based image retrieval using this low level
feature and high level feature is not so impressive.
• Relevance feedback technique is used for refining
queries but it degrades the throughput of the
system, so for this we have to develop new
technique or strategy.
• The performance of these techniques is challenged
by various factors like image resolution, intra-
image illumination variations, non homogeneity of
intra-region and inter-region textures, multiple and
occluded objects.

Related work in relevance feedback
• In 2001, Guo, et al. [9], performed a comparison
between AdaBoost and SVM classifier in content
based image retrieval and found that SVM gives
superior retrieval results.
• In 2004, Tieu and viola [11], proposed a method
for applying the AdaBoost learning algorithm and
noted that it is quite suitable for relevance
feedback due to the fact that AdaBoost works well
with small training sets.

• In 2005, Steven C. H. hoi and Michael R. Lyu
had integrated the user feedback log into
relevance feedback by coupled SVM for CBIR
[14]. By the help of coupled SVM, we learn
consistently with two types of information:
low level image feature and user feedback.
This approach is different from traditional
Relevance feedback approach in CBIR.
Traditionally, in relevance feedback we much
concentrate on low level feature only.

• In 2012, Yuki Sugiyama and Mokoto P. Kato [5] had proposed
the relative relevance feedback in image retrieval that aims to
capture the relative positive and negative examples from the
database for query image concern and then apply the query
refining method such query point movement (Rocchio’s
formula).
=

Here,
= Modified Query vector.
= Original query vector.
= Related document vector.
= Non-related document vector.
= Original Query Weight.
= Related Documents Weight.
= Non-Related Documents Weight.
= Set of Related Documents.
= Set of Non-Related Document.

Soft computing techniques
• There are some soft computing technique by which we
can trained our data for short time perspective or long
term perspective so that we can improve the efficiency
of system.
• There are following type of soft computing technique.
1. Neural network: it is an approach to make the system
like human brain it is used to train the data.
2. Genetic algorithm: it belongs to the evolutionary
algorithm, which generates the solution to optimize
problem using technique inspired by natural evolution
like crossover, mutation, selection and inheritance.

3. Fuzzy based clustering: it is used to make the
system reliable and helpful to achieve the
higher system retrieval performance.
4. Neural network with fuzzy based clustering:
Neuron-fuzzy clustering is used to achieve
the soft computing. Here we first clustered
the image according to fuzzy and then we
train database.

Performance measurement
• There are two parameter on the basis of that
we can find the efficiency of the system.
1. Precision=
2. Recall=

Conclusion
• Relevance feedback is a technique which is using
to get the most appropriate image from large
source of data of image according to our query.
we can refine our output by RF. Relevance
feedback is wide research area on which we can
enhance our results. It is most helpful technique to
reduce the semantic gap between the low level
feature and high level feature but it degrades the
system retrieval efficiency, so for that purpose we
are using some mining technique and soft
computing technique.

References
1. Avneet Kaur Chhabra, Vijay Birchha “A Comprehensive
Survey of Modern Content Based Image Retrieval
Techniques” (IJCSIT) International Journal of Computer
Science and Information Technologies, Vol. 5 (5), 2014,
6127-6129.
2. Tieu and viola “A Comparison and Analysis of Soft
Computing Techniques for Content based Image Retrieval
System” International Journal of Computer Applications
(0975 – 8887) Volume 59– No.13, December 2012
3. Steven C.H. Hoi, Michael R. Lyu “Integrating user
feedback log into relevance feedback by coupled SVM
for content based image retrieval” International
conference on Data engineering.

4. Yuki Sugiyama, Makoto P. Kato and Hiroaki “Relative
relevance feedback in image retrieval” IEEE international
conference on Multimedia and Expo 2012

CBIR with RF

More Related Content

What's hot (20)

Viewers also liked (12)

Similar to CBIR with RF (20)

Recently uploaded (20)

CBIR with RF