IRJET- A Survey on Different Image Retrieval Techniques

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 10 | Oct 2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 1924
A Survey on Different Image Retrieval Techniques
Nandkumar Sushen Admile1
1Karmayogi Polytechnic College, Department of Electronics and Tele-Communication Engineering
Shelve, Pandharpur
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract – Identification of images from animage databaseis
an important research area in image processing. For the
image retrieval number of techniquesare developed. Inearlier
research the text based image retrieval is used, but due to the
number of problems content based image retrieval technique
is proposed. This Paper represents the various image retrieval
techniques. The content based image retrieval is the most
effective technique for image retrieval. In this the word
content refers to the various content of the image such as
color, Shape and Textures. Now a day the block truncation
coding technique is preferred for image retrieval. Out of this
dot diffusion block truncation coding techniqueisemployed in
this paper. To retrieve image using dot diffusion block
truncation two feature descriptor are extracted which are
color histogram feature and bit pattern feature. The
performance of image retrieval task is measured in terms of
average precision rate and average recall rate.
Key Words: Image Retrieval, Content Based Image
Retrieval, Block Truncation Coding, Dot Diffusion Block
Truncation Coding, Average Precision Rate, Average
Recall Rate.
1. INTRODUCTION
Content-based Image Retrieval (CBIR) is also known as
content-based visual information retrieval (CBVIR) or query
by image content (QBIC). It is an application of computer
vision technique to solve the problemofimageretrieval.That
is the problem of searching particular images from a vast
image dataset. Content-based image retrieval is different
from traditional concept-based techniques. The search of
images based on the contents of images instead of metadata
such as keywords, tags, or descriptions belonging to the
images which is the exact meaning of "Content-based". The
term "content" in this case refer to color, shape, texture, or
any other information thatcanbeobtainedfromtheimage.In
most of the systems the search of image totally based on
metadata hence CBIR is suitable for the image retrieval task.
It is too difficult for humans to manually annotate images by
entering keywords or metadata in a vast dataset. It also time
consuming and may not capture the exact keywords to
describe the image. The CBIR systems have lot of challenges
to define the image retrieval task. Now a day’s CBIR is the
most attractive technique for image retrieval because of the
limitations in metadata based systems, as well as the large
range of possible uses for image retrieval. The informationof
textures associated with images canbeeasilyidentifiedusing
existing techniques, but this requiresmanpowertomanually
describe each image in the dataset. It is not easy task for very
large dataset or for images which are generated
automatically, e.g. those from surveillance cameras. There is
alsochance tomissimageswhichusesdifferentsynonymsfor
description .Systems based on classification of images in
semantic classes like "rose" as a subclass of "flowers" can
avoid the mismatching problem, but for this we need more
effort by a user to find images that might be "rose", but are
only classified as an "flowers". Number of techniques have
been developed to categorize images, but all still having
problem of scaling and miscategorization images. The basic
CBIR systems were designed to search dataset based on
image properties such as color, texture, and shape. Content-
based image retrieval, uses the visible contents of an image
such as color, texture, shape and spatial information to
describe the images. In typical content-based imageretrieval
systems (Figure 1-1), the visual contents of the images in the
dataset are obtained and described by multi-dimensional
featurevectors. In order to retrieveimages,userprovidesthe
retrieval system with example images. The system then
changes these examples into its internal representation of
feature vectors. Similarities between the feature vectors of
the query images and those of the images in the dataset are
calculated with the help of indexing scheme and retrieval is
performed. The indexing schemeprovidesaneffectivewayto
search for the images from a vast dataset. Most of the recent
retrieval systems have uses users' relevance feedback to
modify theretrievalprocessinordertogeneratesemantically
and perceptually more accurate retrieval results.
The imagesconsist of various attributes and information.
The main attributes the image consist of following three
things:

1.1 Visual content levels
Natural images consist of various attributes or information
contents that can help to resolvetheimageretrievalproblem.
The information content thatcan be derivedfromanimageis
classified into three levels.
• High level – High level content of images includes
impressions, emotionsandmeaningassociatedwith
the combination of perceptual features. Such as
objects or scenes with emotional or religious
significance.
• Middle level – The examples of middle level contents
include presence or arrangement of specifictypesof
objects, roles and scenes.
• Low level – Low level content of images includes visual
features such as color, texture, shape, spatial
information and motion.
Fig.1.2.Visual content levels of Image
High level features goes beyond the collection of pixels. It
identifiestheimpressions,meaningsandemotionsassociated
with the collection of pixels that make up the object. The
middle level features are thefeaturesthatcanbeextractedby
collection of pixels that make up the image. At the low level,
also called as primary level the features extracted (color,
shape, texture, spatial information and motion) are called
primitive features because they can only extracted from
information obtained at the pixel level.
1.1.1 Color:
Computing distance measures basedoncolorsimilarityis
achieved by computing a color histogram for each image.
Examining images based on the color is one of the most
widely used techniques because it can be completed without
regard to image size or orientation.
1.1.2 Texture
In the field of computer vision and image processing,
there is no clear-cut definition of texture. This is because
available texture definitions are based on texture analysis
methods and the features extracted from the images.
However, texture can be repeated patterns of pixels over a
spatial domain. Texture properties are the visual patterns in
an image that have properties of homogeneity that do not
result from the presence of only a single color or intensity.
The different texture properties as observed by the human
eye for example, regularity, directionality, smoothness, and
coarseness. In real world scenes, texture perception can be
far more complicated. The various brightnessintensitiesgive
rise to a blend of the different human perception of texture.
Image textures have useful applications in image processing
and computer vision. They include: recognition of image
regions using texture properties, known as texture
classification, recognitionoftextureboundariesusingtexture
properties,knownastexturesegmentation,texturesynthesis,
and generation of texture images from known texture
models. Since there is no accepted mathematical definition
for texture, many different methods for computing texture
features have been proposed over the years
1.1.3 Shape:
Shape feature provides the most important information
about an image. Shape features are usually described using
part or region of an image. The accuracy of shape features
depends upon the segmentation usedtodivideanimagewith
more meaningful objects. The shape descriptors are
categorized into two classes:boundary based descriptorand
region based descriptor. Some boundary based
representative shapedescriptiontechniquesarechaincodes,
polygonal approximations, Fourier descriptor and finite
element.
1.2 IMAGE RETRIEVAL METHODS
An image retrievalactive researchareawhichcanbeused
for browsing, searching and retrieving images from a large
dataset of digital images. In this paper following Image
retrieval methods are mentioned.
A.Text Base Image Retrieval
B. Content based Image retrieval
C. Query techniques:
D. Semantic retrieval:
E. Relevance Feedback (Human Interaction):
F. Iterative/Machine Learning:
G. Other query methods:
Text-based retrieval and Content-based retrieval
An image retrievalsystem isa computer basedsystemfor
browsing, searching and retrieving image from a image
dataset. Text-based and content-based are the two
techniques designed to search and retrieve image from
dataset.
In text-based retrieval, images are indexed using
keywords, subject headings or classification codes, which
used as retrieval keysduringsearchandretrieval.Text-based

retrieval isless effective because differentusersusedifferent
keywords to describe images. Text descriptions are
sometimes subjective and incomplete because it cannot
describe complicated image features. Examples are texture
images that cannot be described by text. In text retrieval,
humans are required to describeeachimageinthedataset,so
for a large image dataset the technique is less effective,
expensiveandlabor-intensive.Content-basedimageretrieval
(CBIR) technique uses content of images to search and
retrieve digitalimages.Content-basedimageretrievalsystem
was developed to overcome the limitations of text-based
image retrieval. However, text-based and content-based
image retrieval techniques are complement of each other.
Text-based techniques can capture high-level feature
representation. It is easy to solve text queries but text-based
techniques cannot acceptpicturebasedimagequeries.Onthe
other hand, content-based techniques can capture low-level
image features and accept pictorial queries.
Query techniques:
Different implementations of CBIR make use of different
types of user queries. Query by example is a query technique
which provides theCBIRsystemwithanexampleimages.The
search algorithm may vary depending on the application,but
result images should all share common elements with the
provided example. Options for providing example images to
the system include: A preexisting image may be supplied by
the user or chosen from a random set. The user makes an
approximation of the image that they are looking for, for
example image with color or general shapes. This query
technique removes the difficulties that can existed when
trying to describe images with words.
Semantic retrieval:
Semantic retrieval is nothingbut the usermakingrequest
like "find pictures of Abraham Lincoln". This type of task is
very difficult for computers to perform image retrieval.
Because Lincoln may not always be facing the camera or in
the samepose. Many CBIR systems therefore generallymake
use of lower-level features like texture, color, and shape.
These features are either usedincombinationwithinterfaces
that allow easier input or with datasetthathavealreadybeen
trained to match features (such as faces, fingerprints, or
shape matching). However, in general, image retrieval
requires human feedback in order to identify higher-level
concepts.
Relevance Feedback (Human Interaction):
Combining CBIR techniques is an very difficult task with
the wide range of potential users. In order to make the CBIR
successful depends entirely on the ability of users
understanding. CBIR systems can make use of relevance
feedback, where the user makes the search images in the
resultsas "relevant", "not relevant",or"neutral"tothesearch
query, then repeating the search with the new information.
Iterative/Machine Learning:
Machine learning and application of iterative techniques
are becoming more common in CBIR.
Other query methods:
Along with above mentioned techniques Other query
techniques includebrowsing forexample images, navigating
customized/hierarchical categories, querying by image
region (rather than the entire image), querying by multiple
exampleimages,queryingbyvisualsketch,queryingbydirect
specification of image features, and multimodal queries (e.g.
combining touch, voice, etc.)
LITERATURE REVIEW ON BLOCK TRUNCATION CODING
BASED TECHNIQUE
In 1979 Delp and Mitchell develops the Block Truncation
Coding (BTC) which is for compression of images [1].The
block truncation coding method split the image into
noncontiguous image block. From each block high and low
mean values are calculated. By performing thresholdinglow
mean values bitmap image is derived. In [3] and [5] author
obtains better accuracy using block truncation coding based
image retrieval. In this paper author uses RGB color spaceto
extract the several of image features. In [6] instead of RGB
color space author uses the different color space to extract
the image features. The author obtains theimproved results.
In [2],[4] and [7] BCCM and BPH features are derived to
obtain the sameness between images.In[7]authorusesgray
scale image to perform theimageretrieval task whichproves
it achieves better image quality and efficiency as compared
to ODBTC [8] and EDBTC [9] .In [10] and [11] author uses
DDBTC technique to perform the image retrieval task.
This paper describes the Dot Diffusion Block Truncation
Coding (DDBTC) method for image retrieval. Color
Histogram Features and Bit Pattern Features are extracted
and calculate the results.
Dot Diffusion Block Truncation Coding
In the DDBTC technique a colorimageisdividedinto number
of non-overlapping image blocks. The DDBTC encoder
generates two quantizes (minimumandmaximum)from the
RGB color space (12).
Consider f (i, j) be the image block at position (i, j). Where i =
1, 2….......M/m and j = 1,2..…...N/n.
Let fR (x, y), fG (x, y) and fB (x, y) are the RGB pixel values in
the image block (i,j) where x = 1,2.......m and y = 1,2......n.
Qmin = {min fR (x, y), min fG (x, y) ,min fB (x, y)} (1)
Qmax = {max fR (x, y) ,max fG (x, y) and max fB (x, y)} (2)
R, Gand B represents theRed, Greenandbluecolorspace.
The gray scale image is obtained using:
f^ (x,y) = 1/3 [fR (x, y) + fG (x, y) + fB (x, y)] (3)
From the gray scale image Bitmap image is obtained as:

bm = 1; if f^ (x,y)  f (i,j) (4)
0; if f^ (x,y) < f (i,j)
Fig.1.3 Flow of DDBTC Technique
2. Color Histogram Feature (CHF) :
Using minimum and maximum quantizers ColorHistogram
Feature is extracted The CHF feature is similar to the color
codebook size Nc.
Let C = {C1, C2…. CNc} be the color codebook. Nc the VQ
indexestheDDBTCminimumandmaximumquantizersusing
the symbol I and j.
(5)
(6)
i=1,2,…M/m; j=1,2,….N/n
CHFmin and CHFmax can be calculated as:
(7)
Where k=1,2,……Nmin
(8)
k=1, 2 …Nmax
The Extraction of CHF is as shown in “Fig 2”.
Fig.1.4 Extraction of CHF
3. Bitmap Pattern Feature (BPF):
The Extraction of CHF is as shown in “Fig.4”.
Fig.1.5 Extraction of BPF

Performance Measures:
The Average Precision rate (APR), Average Recall Rate
(ARR) are measured by following equation.
(12)
(13)
The Number of retrieved images is represented by L, total
number of images in database are represented by Nt and
number of relevant images of each class is represented byNr
The query image and correctly retrieved images is
represented by q and Nq respectively.
3. CONCLUSION
In This paper different image retrieval techniques are
studied. For this a 12 number of papers was studied and
various image retrieval techniques and their types and
methods are mentioned such as the text based and content
based and the Semantic based image retrieval. Along with a
DDBTC technique can also employed in this paper.
REFERENCES
[1] E. J. Delp and O. R. Mitchell, “Image coding using
block truncation Coding,” IEEE Trans. Comm., vol. COM-
27, no. 9, pp. 1335–1342, Sep. 1979.
[2] G. Qiu, “Color image indexing using BTC,” IEEE
Trans. Image Process. vol. 12, no. 1, pp. 93–101, Jan.
2003.
[3] S. Silakari, M. Motwani, and M. Maheshwari, “Color
image clustering using block truncation algorithm,” Int.
J. Comput. Sci. Issues, vol. 4, no. 2, pp. 31–35, 2009.
[4] M. R. Gahroudi and M. R. Sarshar, “Image retrieval
based on texture and color method in BTC-VQ
compressed domain,” in Proc. Int. Sym. Signal Process.
Its Appl., Feb. 20, 2007, pp.1–4.
[5] F. X. Yu, H. Luo, and Z. M. Lu, “Colour image retrieval
using pattern co-occurrence matrices based onBTCand
VQ,” Electron. Lett. vol. 47, no. 2, Jan. 20, 2011.
[6] W. Xing Yuan and W. Zongyu, “A novel method for
image retrieval based on structure element’s
descriptor,” J. Vis. Commun. Image Representation, vol.
24, no. 1, pp. 63–74, 2013.
[7] J. M. Guo and Y. F. Liu, “Improved block truncation
coding using optimizeddotdiffusion,”IEEETrans.Image
Process., vol. 23, no. 3,pp. 1269–1275, Mar. 2014.
[8] Jing Ming Guo and Heri Prasetyo, “Content Based
Image Retrieval Using Feature Extracted From Half
toning Based BTC,” IEEE Trans. Vol.24, no.3,pp 1010-
1024,Nov.2014
[9] Jing Ming Guo and Heri Prasetyo, “Effective Image
Retrieval Using Feature Extracted Using Dot Diffused
BTC,” IEEE Trans. Vol.24, no.3, pp 1010-1024,2014
[10] Nandkumar S. Admile and Prof. Rekha Dhawan,
Content Based Image Retrieval Using Feature Extracted
from Dot Diffusion Block Truncation coding, IEEE
International conference on communication and
Electronics Systems (ICCES-2016) Coimbatore, India,
21-22 October 2016.
[11] Nandkumar S. Admile Image Retrieval Based on
Block Truncation coding, IEEE International conference
on communication and Electronics Systems
(ICCES-2018)Coimbatore, India, 15-16 October 2018.

IRJET- A Survey on Different Image Retrieval Techniques

Recommended

Recommended

More Related Content

What's hot

What's hot (17)

Similar to IRJET- A Survey on Different Image Retrieval Techniques

Similar to IRJET- A Survey on Different Image Retrieval Techniques (20)

More from IRJET Journal

More from IRJET Journal (20)

Recently uploaded

Recently uploaded (20)

IRJET- A Survey on Different Image Retrieval Techniques