International Journal of Power Control and Computation(IJPCSC)

Vol 6. No.1 – Jan-March 2014 Pp. 13-17

©gopalax Journals, Singapore
available at : [Link]
ISSN: 0976-268X

CONTENT BASED IMAGE RETRIEVAL USING IMAGE FEATURE

[Link] 2.M. Sunganya [Link] 4. [Link]

Department of computer application
Sathyabamauniversity, chennai 600112.
[Link]@[Link].ms91@[Link] 3. jesudossa@[Link]

ABSTRACT

Humans’ using the images for their communication is the novel process. Images involve in many fields like
medicine, journalism, education and in other areas. The relevance feedback approach to image retrieval is a
powerful technique and has been an active research direction for the past few years. Various ad hoc parameter
estimation techniques have been proposed for relevance feedback. In addition, methods that perform optimization on
multi-level image content model have been formulated. However, these methods only perform relevance feedback
on the low-level image features and fail to address the images’ semantic content. In this proposed mechanism, we
are using the CBIR algorithm for retrieving the images which are related to the input image. Here we are taking the
texture, color and shape of the image and stored in the database. When the user asks query, then it will be matched
with the database and retrieve the image. It will retrieve the exact image by comparing the texture, color and shape.

Key words: Image Retrieval, Data Mining, Relevance Feedback, Content Model, Multilevel Image.

1. INTRODUCTION the relevance feedback system will be able to return

With the increasing availability of digital reasonably accurate results.
images, automatic image retrieval tools provide an
efficient means for users to navigate through them. To address the limitations of the current
Even though traditional methods allow the user to post relevance feedback systems, we propose a framework
queries and obtain results, the retrieval accuracy is that performs relevance feedback on both the images’
severely limited because of the inherent complexity of semantic contents represented by keywords and the
the images for users’ to describe exactly [1]. The more low-level feature vectors such as colour, texture, and
recent relevance feedback approach, on the other hand, shape. The contribution of our work is twofold [3] [4].
reduces the needs for a user to provide accurate First, it introduces a method to construct a semantic
network on top of an image database and uses a simple
Initial queries by estimating the user’s ideal machine learning technique to learn from user queries
query using the positive and negative examples given and feedbacks to further improve this semantic
by the user. The current relevance feedback based network. In addition, we propose a framework in which
systems estimate the ideal query parameters on only semantic and low-level feature based relevance
the low-level image features such as colour, texture, feedback can be seamlessly integrated.
and shape. These systems work well if the feature
vectors can capture the essence of the query. For 2. EXISTING TECHNIQUES
example, if the user is searching for an image with A symbolic image is an array representing a
complex textures having a particular combination of set of objects and a set of spatial relations among them.
colours, this query would be extremely difficult to Symbolic images and related structures have been used
describe but can be reasonably represented by a in a number of applications including Image Databases,
combination of colour and texture features[2][14]. Spatial Reasoning, Path Planning and Spatial Pattern
Therefore, with a few positive and negative examples, Matching. In this paper we describe a Pictorial Query-

gopalax Publications 13
 International Journal of Power Control and Computation(IJPCSC)
Vol 6. No.1 – Jan-March 2014 Pp. 13-17
©gopalax Journals, Singapore
available at : [Link]
ISSN: 0976-268X

By-Example (PQBE) language aimed at the retrieval of 4.1. Image pre-processing

direction relations from symbolic images. As in the 4.2. Feature extraction
case of verbal Query-By-Example, PQBE generalises 4.3. Training of images
from the example given by the user, but instead of 4.4. Retrieval of Image
having queries in the form of skeleton tables showing
the relation scheme, we have skeleton images which 4.1Image Pre-Processing
are they symbolic images [5]. In this article, we give an
overview of the main tasks involved in designing a The content retrieval
platform for the evaluation of content-based image is based on the Image Pre-processing module [8] [9].
retrieval systems. A number of issues should be Here, an image will be stored in the database using the
addressed from the construction of an image collection three level processes. They are:
to the definition of standard performance measures. 1. Low level process:It is done for reducing the noise in
the image.
The relevance feedback approach to image 2. Mid level process:It is done for segmenting and
retrieval is a powerful technique and has been an active describing the image.
research direction for the past few years [15]however,
these methods only perform relevance feedback on the 3. High Level process: It is done for making sense to
low-level image features and fail to address the human vision.
images’ semantic content. In this paper, we propose a
relevance feedback technique, iFind, to take advantage Set x, y coordinates
of the semantic contents of the images in addition to and amplitude f to
the low-level features. By forming a semantic network
on top of the keyword association on the images, we
are able to accurately deduce and utilize the images’ Image is digitized
semantic contents for retrieval purposes [6]. The
accuracy and effectiveness of our method is
demonstrated with experimental results on real-world Three- level process
image collections. is done

3. CONTENT BASED IMAGE RETRIEVAL

SYSTEMS Low-level to reduce
Content-based image retrieval (CBIR), also
known as query by image content (QBIC) and content-
based visual information retrieval (CBVIR) is the Mid-level to segment
and Description
application of computer vision to the image retrieval about image
problem. This system is based upon a combination of
higher-level and lower-level vision principles [16].
Higher-level analysis uses perceptual organization, High- level to making
inference and grouping principles to extract semantic sense to objects
information describing the structural content of an
image. Lower-level analysis uses image texture, shape Figure .1Image pre-processing
and color histogram techniques. The image search is of
two types such as target search and category search. 4.2 Feature Extraction
The goal of target search is to retrieve known and Here, the features of an image are analyzed and
specific image, such as registered logo, a historical stored in the database. The features are Texture, Shape
photograph, or a particular painting [7]. The goal of and Colour.
category search is to retrieve a given semantic class or 1. In Texture analysis the texture features are
genre of images or used to find relevant images that the extracted
user might not be aware ahead of time, such as scenery
images or skyscrapers. 2. In Shape analysis the shape will be determined by
considering the Edge.
4 IMAGE PROCESSING FUNCTIONS Here, the edges of the image are taken. Thus, the
shape is considered

gopalax Publications 14
 International Journal of Power Control and Computation(IJPCSC)
Vol 6. No.1 – Jan-March 2014 Pp. 13-17
©gopalax Journals, Singapore
available at : [Link]
ISSN: 0976-268X

1. First, the image is given as the input to the from the

3. The last is Colour analysis. The RGB values are camera.
considered here. Thus, the colour features are 2. Initially, this image is a raw image where it contains
extracted. noise.
3. Then, the Features like Texture, Color and Shape are
extracted by the RGB values
4. These features values and database values are
matched. If it matched there will be fast retrieval of the
image is done. The content of the image is also
retrieved.

Figure2. Feature extraction

4.3 Training of Images

After extracting the image features like
texture and matrix conversion the pixel values are
trained in the database by labelling the features of the
images. The matrix conversion is done by giving
intensity at each point x, y and RGB values are found
[12] [13]. A matrix will be formed having M rows and
N columns. Then the images are labelled in the
database. So, it can be retrieved from the database
easily. These labelling is done by the features of the
image. Now, the image is stored in the database.

Figure. 4Image retrieval

5. CONCLUSION
In this paper, a new framework is presented in
which semantics and low-level feature based relevance
feedbacks are combined to help each other in achieving
higher retrieval accuracy with lesser number of
feedback iterations required from the user. The novel
feature that distinguished the proposed framework
from the existing feedback approaches in image
database is twofold. First, it introduces a method to
Figure .3Training of image
construct a semantic network on top of an image
database and uses a simple machine learning technique
to learn from user queries and feedbacks to further
4.4 Image Retrieval
improve this semantic network. In addition, a scheme
Now, the image is given as the input image for
is introduced in which semantic and low-level feature
extracting the [Link] results are extracted by the
based relevance feedback is seamlessly Experimental
following process:
evaluations of the proposed framework have shown

gopalax Publications 15
 International Journal of Power Control and Computation(IJPCSC)
Vol 6. No.1 – Jan-March 2014 Pp. 13-17
©gopalax Journals, Singapore
available at : [Link]
ISSN: 0976-268X

that it is effective and robust and improves the retrieval

performance of CBIR systems significantly
6. FUTURE ENHANCEMENT
• Histogram of 25 colors and 25 textures for
each image is computed from a vector
quantization.
• It is based on precision and recall. The
precision and recall are considered for one
category and have as many values as image in
the database.
• Mainly used for medical purposes.

7. EXPERIMENTAL RESULTS
Figure8. Extracted image

[Link] the Disease Name and Browse the

Image ofthe Disease Figure9. Texture and Shape. Image Is Shown

Figure 6 . RGB Extration

Figure10. Image Is Stored

REFERENCES
[1]T. Joachims, “Transductive inference for text
classification using support vector machines,” in Proc.
16th International Conference on Machine Learning.
Morgan Kaufmann, San Francisco, CA, 1999, pp. 200–
209.

Figure 7. Relevant image based on query

gopalax Publications 16
 International Journal of Power Control and Computation(IJPCSC)
Vol 6. No.1 – Jan-March 2014 Pp. 13-17
©gopalax Journals, Singapore
available at : [Link]
ISSN: 0976-268X

[2]R. Veltkamp, “Content-based image retrieval Modeling and Relevance Feedback Approach for
system: A survey,” University Content-Based Image Retrieval”, Systems, Signals and
of Utrecht, Tech. Rep., 2002. Image Processing, 2009,Page 1-5.
[11]. Y Lu , C Hu, X Zhu et al1 A unified semantics
[3]E. Chang, B. T. Li, G. Wu, and K. Goh, “Statistical and feature based image retrieval technique using
learning for effective visual information retrieval,” in relevance feedback1 In : Proc of the 8th ACM Int’l.
IEEE International Conference on Image Processing, Multimedia Conf. Los Angeles: ACM Press, 2001.
Barcelona, Spain, September 2003, pp. 609–612. [12]A. Vailaya, Semantic Classification inImage
[4]S. Mika, G. R¨atsch, J. Weston, B. Sch¨olkopf, and Databases, PhD thesis, Department of Computer
K.-R. M¨uller, “Fisher discriminant analysis with Science, Michigan State University, 2000.
kernels,” in Neural Networksfor Signal Processing IX, [13]A. Vailaya, M. Figueiredo, A. K. Jain, and H.J.
Y.-H. Hu, J. Larsen, E. Wilson, and S. Douglas, Eds. Zhang, Image Classification for Content-Based
IEEE, 1999, pp. 41–48 Indexing , IEEE Transactions on Image Processing,
vol. 10, no. 1, pp 117-130, January, 2001.
[5] P. Yin, B. Bhanu, K. Chang, and A. Dong, [14]. Sun Yen , Wang Zheng-xuan , Wang Dong-mei
“Integrating relevance feedback techniques for image “An Image Retrieval Method Based on Relevance
retrieval using reinforcement learning,” IEEE Trans. Feedback and Collaborative Filtering”, 2010.
on Pattern Analysis and Machine Intelligence, vol. 27, [15]YihunAlemu, Jong-bin Koh, MuhammedIkram,
pp. 1536–1551, October 2005. Dong-Kyoo Kim “Image Retrieval in Multimedia
Databases: A survey”, Fifth International Conference
[6]Philippe Henri Gosselin and on
matthieucord(1).”content based image retrieval” in [16] P. Gosselin and M. Cord, “RETIN AL: An active
IEEE international conference on image processing learning strategy for image category retrieval,” in IEEE
International Conference on Image Processing, vol. 4,
[7]J. Shawe-Taylor and N. Cristianini, Kernel methods Singapore, October 2004, pp. 2219–2222.
for Pattern Analysis. Cambridge University Press,
ISBN 0-521-81397-2, 2004. [17]. A. Vailaya, M. Figueiredo, A. K. Jain, and H.J.
Zhang, Image Classification for Content-Based
[8] P. Gosselin, M. Najjar, M. Cord, and C. Ambroise, Indexing , IEEE Transactions on Image Processing,
“Discriminative classification vsmodeling methods in vol. 10, no. 1, pp 117-130, January, 2001.
CBIR,” in IEEE Advanced Concepts for Intelligent
Vision Systems (ACIVS), Brussel, Belgium, September
2004.

[10] Z. Barceló’s, E. L. Flores; C. A. Z. Barceló’s, S. F.

Silva, M. A. Batista,” A Multi-Dimensional Similarity

gopalax Publications 17