Upload
others
View
15
Download
0
Embed Size (px)
Citation preview
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8474
Color and Texture Based Image Retrieval Feature Descriptor using Local
Mesh Maximum Edge Co-occurrence Pattern
Harpreet Kaur1 and Vijay Dhir2
1Research Scholar, 2 Professor, 1,2 Department of Computer Science and Engineering,
Punjab Technical University, Kapurthala, Punjab, India.
Abstract
A novel descriptor is designed and developed for extracting
the features from large databases. The descriptor is known as
LMeMECoP that is local mesh maximum edge co-occurrence
patterns. Information of local maximum edge is collected
between the possible neighbors for provided centre pixel that
is the reference pixel. The maximum edge collection is in the
form of mesh intended for reference pixel. The information of
extracted maximum edge is utilized for forming binary bits.
The co-occurrence bits are collected later for appropriate
centre pixel by means of maximum edge response
information. In the end, the patterns are generated from the
co-occurrence bits between the provided pixels with its
neighbor pixels. The combination of texture information with
color information takes place for enhancing the proposed
method performance. The change of RGB to HSV color space
is executed for general histogram and color centered feature
for each color space. Generation of final feature is executed
with the integration of texture based features with color based
features. The LMeMECoP assessment is executed by
proposed method testing on Corel 5K with Corel 10 K and
MIT VisTex dB centered on ARP (average retrieval
precision). The output after investigation has shown that
LMeMECoP has shown a tremendous betterment in terms of
ARP on Corel-5K and Corel-10K dB.
Keywords: Local mesh patterns, maximum edge patterns;
Pattern Recognition; texture; image retrieval
INTRODUCTION
Image Retrieval is known as a method for retrieving the
images from huge image databases as user’s requirements. It
is intended for searching the image from the actual content
instead of metadata. The concept of image retrieval is utilized
for extracting the features initially and then indexing them by
utilizing suitable structures and effectively executes the
answer as per the user’s query. Image Retrieval firstly takes
the RGB image in the form of Input, operates feature
extraction with few of its similarity computations by the
images being stores in the databases and later retrieves the
output image as per similarity computation. Some general
Image retrieval fundamentals are there and are categorized
into three parts, namely, feature extraction, multi dimensional
indexing and architecture of retrieval system. The diagram for
the same is shown in figure 1.
The objective of the research is the designing of effective and
competitive algorithm for visual features and combination of
similar one. The improvements in the field are shown by
number of researchers and are provided in [1-4].
Figure 1: Image Retrieval Block Diagram
Query Image
Feature Extraction
Database Image
Similarity Evaluation
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8475
Texture calculates the image characteristics as per the
variation in definite scale and directions [5-7]. It basically
extracts the significant information of surface structural
arrangement like fabrics, clouds and bricks, so on and
association with the relative environment. Generally, the
textural information examines the pixels group behavior
instead of single pixel nature. The natural images are
considered as the premium examples of texture and color
mixture. While number of texture descriptors performs on
gray space but for optimizing the feature performance, they
may perform on color images. Varied types of strategies are
followed for extracting the texture features [8].
The concept of Wavelet Transform based correlogram for
image retrieval was given by Moghaddam et al. [9]. The
wavelet correlogram efficiency is enhanced by utilizing
metaheuristic GA (Genetic Algorithm) [10] for the
optimization of quantization thresholds. The color histogram
(color) and wavelet transform (texture) features are integrated
by Birgale et al. [11] with Subrahmanyam et al. [12] for CBIR
(Content Based Image Retrieval). The Correlogram algorithm
for the retrieval of images has utilized WC+RWC (wavelets
and rotated wavelets) and is proposed in [13].
LBP (Local binary pattern) is being proposed for defining the
textures by Ojala et al in [14]. The features for LBP was
proposed by Ojala et al. are dependent on rotational variant.
The rotation variant based LBP is transformed into rotational
invariant by means of texture classification [15-16]. For facial
expression recognition and analysis, the LBP features are
utilized in [17, 18]. Heikkila et al. has presented the detection
and modeling background by utilizing LBP in [19]. Huang et
al. has presented the comprehensive LBP for shape
localization in [20]. Heikkila et al. has utilized the LBP for the
description in interest region in [21]. Li et al. has utilized the
integration of LBP and Gabor filter for the segmentation of
texture in [22]. Zhang et al. has defined LDP (Local derivative
pattern) for face recognition in [23]. The authors have taken
LBP as a local pattern of non-directional first order that are
the binary outcome of the derivate of first-order in images.
Block-based LBP texture feature is utilized as image
description basis for Content Based Image Retrieval in [24].
An enhanced version of the LBP features is named as CS-LBP
(Center-symmetric local binary pattern) is connected with
SIFT (Scale invariant feature transform) for defining the
interest regions in [25]. Two varied type of LEP (Local edge
pattern) histograms, namely, LEPINV and LEPSEG are given
by Yao et al. in [26]. LEPSEG is given for image and
LEPINV segmentation and is designed for the retrieval of
images. The LEPSEG is scale and rotation variant, but, the
LEPINV is scale and rotation invariant.
The LBP and the LDP addition in the text cannot deal with the
different exterior discrepancies that mainly occur in
unrestrained natural images because of pose, aging, facial
expression, illumination, partial obstructions, and so on.
Therefore, for solve the defined problem; LTP (Local ternary
pattern) has been defined for face recognition in diverse
lighting environment in [27]. Few researches on pattern
dependent features by Subrahmanyam et al. has, LMEBP
(local maximum edge patterns) in [28], LTrP (local tetra
patterns) in [29] and DLEP (directional local extrema
patterns) in [30] for natural as well as texture image retrieval
and DBWP (directional binary wavelet patterns) in [31],
LMeP (local mesh patterns) in [32] and LTCoP (local ternary
co-occurrence patterns) in [33] for retrieval of medical
images. Reddy et al. in [34] has enhanced the features of
DLEP by computing the image local gray values of magnitude
information. Jacob et al. in [35] has given LOCTP (local
oppugnant color texture pattern) for retrieving the images.
Different existing features with different ways are being
provided in the literature for gathering the connection among
the referenced pixels with its neighbors between the neighbors
for applications of pattern recognition.
These mentioned features have inspired us to present
LMeMECoP (Local mesh maximum edge co-occurrence
patterns) for the applications of image retrieval. The methods
connect the information of maximum edge between neighbors
and then compile the maximum edge bits co-occurrence and
are coded as patterns. The algorithm being proposed is more
contrasting as compare to existing algorithms. Further, the
proposed method performance is enhanced by integrating it
through color features. The proposed method evaluation is
implemented on image database of benchmark Corel-10K
(10000). The paper is organized in the following manner.
A concise summary of image retrieval with its glance is
defined in Section I. the review of existing local feature
descriptors is defined in Section II. The proposed feature
descriptor is elaborated in Section III. The results obtained
after the simulation of the proposed work is defined in
Section IV. The conclusion and the future scope are being
drawn on the basis of simulated work in Section V.
REVIEW OF EXISTING LOCAL FEATURE
DESCRIPTORS
LBP (Local Binary Patterns)
The concept of LBP was defined by Ojala et al for texture
classification [14]. This method gives a robust way to define
exact local patterns in the texture. The exact 3X3
neighborhood threshold values are defined by the center
pixels that are added by binomial values for the equivalent
pixels. The pixels being resulted is added for LBP number for
the mentioned texture unit. The method LBP is known as a
gray scale invariant and could be simply integrated with
contrast measure by calculating the average gray level
difference of every neighborhood with value 1 and 0
correspondingly as shown in figure 2.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8476
Figure 2: LBP Block Diagram
Local binary pattern is known as a two valued code. The
values of LBP are calculated with the comparison of centre
pixel’s gray value with the neighbors by utilizing the below
equations:
(1)
(2)
From the above equation,
= Centre pixel’s gray value
Neighbors’s gray value
Q=Number of neighbors
R= Neighborhood radius
Figure 3: LBP Value computation
Original Image (3x3)
pixels
Output Image
6 5 2
7 6 1
9 3 7
8 0 0
16 0
32 0 128
1 0 0
1 0
1 0 1
8 4 2
16 1
32 64 128
6 5 2
7 6 1
9 3 7
1 0 0
1 0
1 1 1
1 0 0
128 0
64 32 16
241
Binary Pattern Example
Weights LBP Value
LBP=1+16+32+64+128=241
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8477
Local Maximum edge Binary Patterns (LMEBP)
LMEBP operator gathers the information of maximum edge
on the basis of local difference magnitude among the
referenced pixel with its required neighbours intended in an
image [28]. The calculation of initial maximum edge is
defined below:
where, j=1,2,......8 (3)
(4)
If the corresponding edge sign is positive, than it is coded as
‘1’ or else it is coded as ‘0’ for the source pixel.
(5)
LMEBP can be described as:
(6)
On the basis of above mentioned equation 6, the LMEBP code
is being taken for every source pixel for some image. In the
end, LMEBP histogram is produced for LMEBP resultant map
[28].
Figure 4: Local Maximum edge Binary Patterns
Local Mesh Patterns (LMeP)
Local Mesh patterns were proposed by Subrahmanyam et al in
[38] for image indexing and image retrieval. It elaborates the
structure of local spatial for texture by utilizing the
relationship among nearest neighbors for provided center
pixel. The local mesh patterns for matrix of 3X3 is achieved
by computing differences among two surroundings pixels for
the centre pixel as defined below:
(7)
As shown in figure 7,j=1,2,…,8 and +((i+Q+i-1) mod P)
(8)
As defined above, j is showing initial three patterns, s defines
the nearest neighbors within the centre pixel. When the LMeP
calculation takes place, the image is shown by calculating the
histogram as shown in below figure:
(9)
, in which M and N are defined as image
size.
Figure 5: Local Mesh Patterns
Local Maximum Edge Co-occurrence patterns
(LMECoPs)
For some provided image, initially LMEBP patterns are
integrated and then the cooccurrence matrix is being taken on
the LMENP image. The provided Co-occurrence matrix is
collected with HSV color histogram [39]. The working of
Local maximum Edge Co-occurrence patterns is defined
below:
Initially, RGB image is loaded and is being converted into
HSV color space for the extraction of color features.
Construction of H,S and V for space histogram is taken place.
Later, the RGB image is converted into gray scale for the
extraction of text features. The maximum edge information
among the center pixel with its neighbors is collected and then
the maximum edge patterns are calculated. A feature vector is
formed by integrating the HSV histogram with Co-occurrence
matrix. The best matches on the basis of similarity measures
between the target image and the query image are calculated
from the databases for retrieving the fine matches.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8478
Proposed Feature Descriptor
Local Mesh Maximum Edge Co-occurrence Patterns
(LMeMECoP)
The proposed method that is LMeMECoP gathers the
relationship between the neighbors for the calculation of
maximum edge being dissimilar from LMEBP. Firstly, the
local dissimilarity between the neighbors for provided source
pixel is defined below:
, where i=1, 2… Q; (9)
((i+Q+k-1), Q);
(10)
As shown above, r is the neighbor radius by means of center
pixel, Q is the number of neighbors, k is the local dissimilarity
index. The mod (x,y) proceeds with the reminder for operation
x and y.
Maximum edges are composed for appropriate k
(11)
When the sign of equivalent edge is taken as positive, than it
is considered as 1 instead of 0 for the source pixel
(12)
LMeME that is Local mesh maximum edge bits are integrated
for a provided centre pixel for appropriate i as defined below:
(13)
In the end, the pattern of co-occurrence is taken for provided
center pixel at appropriate i is defined below:
………. (14)
In the above equation (14), x*y is the pattern co-occurrence
The whole needed operators of LMeMeCoP for some
provided image is defined below:
Finally, the provided image is being changed into
LMeMECoP maps with values from 0 to 255. After
calculating the LMeMECoP, the entire map is shown by
developing a histogram as defined by below equation:
(15)
Similarity measure and query matching
The feature vector intended for query image I is defined as
(16)
The above equation is resulted after the extraction of features.
Likewise, every image intended in the database is shown with
feature vector
(17)
In the above equation, i=1, 2…
The aim is to choose an m best image which is same as the
query image. It generally consists of m mostly matched image
by calculating the distance among query image and database
image. For matching the image, we have utilized similarity
distance metric as calculated by below equation [30]:
(18)
As shown in the above equation (18), is the kth feature
of ith image in database.
Image Retrieval System framework
This research has presented a novel feature descriptor,
LMeMECoP (local mesh maximum edge co-occurrence
patterns) and combines with the color features with Indexing
of color based image with the retrieval applications.
The algorithm of the proposed image retrieval system is
defined below:
Proposed Image Retrieval Algorithm
In the below algorithm, Input is the Image and output is the
retrieval result.
Step 1: The color image is loaded and is being changed into
gray scale (g) with HSV value.
Step 2: For some specified center pixel, gather the neighbors
from gray scale image.
Step 3: Compute the local dissimilarity between the
neighbors.
Step 4: Gather the maximum edge bit by means of local
difference and code the bit as 1 and 0 on the basis of
sign.
Step 5: Employ steps 2 to 4 for every pixels by taking all
pixels as source pixel.
Step 6: Describe the co-occurrence pattern by utilizing
equation 14.
Step 7: Develop the histogram from LMeMECoP with
dissimilarity k.
Step 8: Compute HSV histogram for S,H and V color image
space.
Step 9: Develop the feature vector by integrating all
histograms.
Step 10: Evaluate the query image with the image intended in
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8479
the dB by utilizing equation 18
Step 11: Recover the image on the basis of best matches.
EXPERIMENTAL RESULT
The effectiveness of the work being proposed is tested by
executing the two experiments on benchmark databases. The
databases utilized for the implementation are:
i. Core-5K
ii. Corel-10
iii. MIT VisTex database
For experiments, 1 and 2, Corel database has been used for
images [36]. The Corel database is consisted of variety of
images for different content ranges from animals to outdoor
sports for natural images. The mentioned images are
categorized into dissimilar categories with size of 100 by area
professionals. Few researchers usually believe that Corel
database fulfill all the requirements for accessing an image
retrieval system because of its huge size and diverse content.
In each experiment, all images in the database are utilized as a
query image. For every query, the system compiles
databases images, with the matching
distance of small images measured by utilizing equation 17.
When the retrieved image is belong to
similar category as per the query image, then it can be
concluded that the system have appropriately recognized the
predictable image or the system is failed for finding the
predictable image.
For the evaluation of the proposed method, performance
metrics, namely, Precision, Average Retrieval Precision
(ARP), Recall and Average Retrieval Recall (ARR) has been
used and are defined below for some query image :
i. Precision= P( =
ii. Average Retrieval Precision= ARP( =
iii. Recall= R( =
iv. Average Retrieval Recall= ARR( =
Database -1: Corel-5K Database
The Corel-5K dB is consisted of 5000 natural images for 50
dissimilar categories. Every category is consisted of 100
images. The 50 categories are the natural images like people,
animal, buses, beaches and so on. The performance for the
proposed image retrieval system is executed on the basis of
precision, recall, ARP and ARR as well.
The results for the existing and the proposed system is
depicted in the Table I on the basis of Corel-5K and Corel-
10K dB’s for average precision and recall ad the performance
has been shown on the basis of different categories is shown
in figure 7(a) and 7(b) correspondingly for Corel-5K dB.
Then, the performance for different methods is calculated by
means of ARP and ARR on Corel 5K dB and the results are
shown in figure 8(a) and 8(b) correspondingly.
Table I: Different methods for Precision and Recall on Corel–5k and Corel–10k databases MDLEP: DLEP+MDLEP; BLK_LBP:
Block based LBP, Proposed method: LMeMECoP
Data
Base
Perfor-
mance
Method
Corel–5K CS_
LBP
LEPSE
G
LEPIN
V
BLK_
LBP
LBP DLE MDLE
LOC
TP
LME
COP
LMe
MECo
Precision (%) 32.9 41.5 35.1 45.7 43.6 48.8 54.4 56.4 58.7 62.2
Corel–10K Recall (%) 14.0 18.3 14.8 20.3 19.2 21.1 24.1 26.1 28.5 30.2
Precision (%) 26.4 34.0 28.9 38.1 37.6 40.0 45.4 47.3 __ 52.9
Recall (%) 10.1 13.8 11.2 15.3 14.9 15.7 18.4 20.3 __ 22.9
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8480
As shown in table 1, figure 7 and figure 8, below points has
been observed:
i. The LCOMeEP defines 29.3%, 27.1%, 20.7%, 18.6%,
16.5%, 7.8%,13.4%, and 5.8% enhanced results as related to
LEPSEG, CS-LBP, BLK_LBP, LEPINV, LBP, MDLEP, and
LOCTP correspondingly for ARP on Core-5K dB.
ii. The LCOMeEP defines 11.9%,16.2% , 9.9%, 15.4%,
11.0%, 6.1%,9.1%, and 4.1% enhanced results as related to
CS-LBP, LEPSEG, LEPINV, BLK_LBP, LBP, DLEP,
MDLEP and LOCTP correspondingly for ARR on Core-5K
dB.
It has been observed that the methods being proposed has
depicted a major improvement as related to state-of-the-art
techniques for the computation measures on Corel–5K dB.
Fig. 8 defines the results of query retrieval of LMeMECoP on
Corel–5K dB.
Database-2: Corel-10K Database
The Corel-10K dB is consisted of 10000 natural images for
100 dissimilar categories. Every category has 100 images. The
proposed image retrieval system performance is calculated on
the basis of precision, recall, ARP and ARR.
The different method performance for average precision and
recall is shown in figure 6 (a) and Fig. 6 (b) correspondingly
on Corel-10K dB. Further, different methods performances
are also calculated on the basis of ARP and ARR on Corel-
10K dB. The ARP and ARR results are shown in Fig. 7(a) and
Fig. 7(b) correspondingly. From Table 1, Figure 6 and Figure
7, it is concluded that the performance of the proposed method
shows an improvement as compare to state-of-the-art methods
on Corel–10K dB.
Figure 6: Comparison of LMeMECoP with different existing methods on Corel–5K.dB (a) Category wise performance for
precision, (b) category wise performance for recall
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8481
Figure 7: Comparison of different methods for ARP and ARR on Corel-5K dB
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8482
Figure 8: Image retrieval example by LMeMECOP on Corel–5K dB
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8483
Figure 9: LMeMECoP comparison with different existing methods on Corel–10K dB (a) Category wise performance for
precision, (b) category wise performance for recall
Figure 10: Various methods comparison for ARP and ARR on Corel-10K dB
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8484
Figure 11: Proposed method Comparison with existing methods for ARR on MIT VisTex dB
Database 3: MIT VisTex Database
MIT VisTex database is taken that contains 40 varied textures
[37]. Every texture has a size of 512X512 being categorized
into sixteen 128X128 non-overlapping sub-images and
therefore, developing 640 (40X16) images database.
The performance of the proposed method that is LMeMECoP
by means and different other methods as a function number as
depicted in figure 10 and it is being shown that the method
being proposed has performed better as compare to other
existing methods for average retrieval rate on MIT VisTex
dB.
CONCLUSION
In this research, novel descriptor, that is, LMeMECoP (local
mesh maximum edge co-occurrence patterns) has been
proposed for feature extraction. The presented descriptor
initially gathers the information of local maximum edge
between the probable neighbors for provided reference pixel
that is the centre pixel. The binary bits are produced later on
from the information of extracted maximum edge. The
proposed method gathers the co-occurrence binary bits from
the provided center pixel by means of the response of
maximum edge information. The method later gathers the co-
occurrence binary bits for specified center pixel of the
response of maximum edge. The performance is being
evaluated on the basis of three databases, namely, Corel-5K,
Corel-10K and MIT VisTex database for precision, recall,
ARP (average retrieval precision) and ARR (average retrieval
rate). The outcome after execution has shown an improvement
as compare to existing work for image retrieval.
ACKNOWLEDGEMENT
One of our authors, Harpreet Kaur wants to thanks to I.K
Gujral Punjab Technical University (PTU), Kapurthala, India
for their help and support.
REFERENCES
[1]. Vipparthi, S. K., & Nagar, S. K. (2014). “ Multi-joint
histogram based modelling for image indexing and
retrieval”,Computers & Electrical Engineering, 40(8),
163-173.
[2]. Subrahmanyam, M., Wu, Q. J., Maheshwari, R. P., &
Balasubramanian, R. (2013). “Modified color motif
co-occurrence matrix for image indexing and
retrieval”, Computers & Electrical Engineering, 39(3),
762-774.
[3]. Fathima, A. A., Vasuhi, S., Babu, N. T., Vaidehi, V., &
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8485
Treesa, T. M. (2014). “Fusion Framework for
Multimodal Biometric Person Authentication
System”, IAENG International Journal of Computer
Science, 41(1).
[4]. Lan, M. H., Xu, J., & Gao, W. (2016). “ Ontology
feature extraction via vector learning algorithm and
applied to similarity measuring and ontology
mapping”, IAENG International Journal of Computer
Science, 43(1), 10-19.
[5]. Poon, B., Amin, M. A., & Yan, H. (2016). “ PCA
Based Human Face Recognition with Improved
Methods for Distorted Images due to Illumination and
Color Background”,IAENG International Journal of
Computer Science, 43(3).
[6]. Xia, Z., Yuan, C., Sun, X., Sun, D., & Lv, R. (2016).
“Combining wavelet transform and LBP related
features for fingerprint liveness detection”, IAENG
International Journal of Computer Science, 43(3), 290-
298.
[7]. Manjunath, B. S., & Ma, W. Y. (1996). “Texture
features for browsing and retrieval of image data”,IEEE
Transactions on pattern analysis and machine
intelligence, 18(8), 837-842.
[8]. Kokare, M., Biswas, P. K., & Chatterji, B. N. (2007).
“Texture image retrieval using rotated wavelet
filters”,Pattern recognition letters, 28(10), 1240-1249.
[9]. Moghaddam, H. A., Khajoie, T. T., & Rouhi, A. H.
(2003, September). “A new algorithm for image
indexing and retrieval using wavelet correlogram.
In Image Processing,” 2003. ICIP 2003. Proceedings.
2003 International Conference on (Vol. 3, pp. III-497).
IEEE.
[10]. Saadatmand, M. T., & Moghaddam, H. A. (2005,
September). “Enhanced wavelet correlogram methods
for image indexing and retrieval”, In Image Processing,
2005. ICIP 2005. IEEE International Conference
on (Vol. 1, pp. I-541). IEEE.
[11]. Birgale, L., Kokare, M., & Doye, D. (2006, July).
“Colour and texture features for content based image
retrieval”, In Computer Graphics, Imaging and
Visualisation, 2006 International Conference on (pp.
146-149). IEEE.
[12]. Murala, S., Gonde, A. B., & Maheshwari, R. P. (2009,
March). “Color and texture features for image indexing
and retrieval”, In Advance Computing Conference,
2009. IACC 2009. IEEE International (pp. 1411-1416).
IEEE.
[13]. Subrahmanyam, M., Maheshwari, R. P., &
Balasubramanian, R. (2011). “A correlogram algorithm
for image indexing and retrieval using wavelet and
rotated wavelet filters”, International Journal of Signal
and Imaging Systems Engineering, 4(1), 27-34.
[14]. Ojala, T., Pietikäinen, M., & Harwood, D. (1996). “A
comparative study of texture measures with
classification based on featured distributions”, Pattern
recognition, 29(1), 51-59.
[15]. Ojala, T., Pietikainen, M., & Maenpaa, T. (2002).
“Multiresolution gray-scale and rotation invariant
texture classification with local binary patterns”, IEEE
Transactions on pattern analysis and machine
intelligence, 24(7), 971-987.
[16]. Pietikäinen, M., Ojala, T., & Xu, Z. (2000). “Rotation-
invariant texture classification using feature
distributions”, Pattern Recognition, 33(1), 43-52.
[17]. Ahonen, T., Hadid, A., & Pietikainen, M. (2006). “Face
description with local binary patterns: Application to
face recognition. IEEE transactions on pattern analysis
and machine intelligence”,28(12), 2037-2041.
[18]. Zhao, G., & Pietikainen, M. (2007). “Dynamic texture
recognition using local binary patterns with an
application to facial expressions. IEEE transactions on
pattern analysis and machine intelligence”, 29(6), 915-
928.
[19]. Heikkila, M., & Pietikainen, M. (2006). “A texture-
based method for modeling the background and
detecting moving objects”,. IEEE transactions on
pattern analysis and machine intelligence, 28(4), 657-
662.
[20]. Huang, X., Li, S. Z., & Wang, Y. (2004, December).
“Shape localization based on statistical method using
extended local binary pattern”, In Image and Graphics
(ICIG'04), Third International Conference on (pp. 184-
187). IEEE.
[21]. Heikkilä, M., Pietikäinen, M., & Schmid, C. (2009).
“Description of interest regions with local binary
patterns”, Pattern recognition, 42(3), 425-436.
[22]. Li, M., & Staunton, R. C. (2008). “Optimum Gabor
filter design and local binary patterns for texture
segmentation”, Pattern Recognition Letters, 29(5), 664-
672.
[23]. Zhang, B., Gao, Y., Zhao, S., & Liu, J. (2010). “ Local
derivative pattern versus local binary pattern: face
recognition with high-order local pattern
descriptor”, IEEE transactions on image
processing, 19(2), 533-544.
[24]. Takala, V., Ahonen, T., & Pietikäinen, M. (2005).
“Block-based methods for image retrieval using local
binary patterns”, Image analysis, 13-181.
[25]. Heikkilä, M., Pietikäinen, M., & Schmid, C. (2009).
Description of interest regions with local binary
patterns. Pattern recognition, 42(3), 425-436.
International Journal of Applied Engineering Research ISSN 0973-4562 Volume 12, Number 19 (2017) pp. 8474-8486
© Research India Publications. http://www.ripublication.com
8486
[26]. Yao, C. H., & Chen, S. Y. (2003). “Retrieval of
translated, rotated and scaled color textures”, Pattern
Recognition, 36(4), 913-929.
[27]. Tan, X., & Triggs, B. (2010). “Enhanced local texture
feature sets for face recognition under difficult lighting
conditions”, IEEE transactions on image
processing, 19(6), 1635-1650.
[28]. Subrahmanyam, M., Maheshwari, R. P., &
Balasubramanian, R. (2012). “Local maximum edge
binary patterns: a new descriptor for image retrieval
and object tracking”, Signal Processing, 92(6), 1467-
1479.
[29]. Murala, S., Maheshwari, R. P., & Balasubramanian, R.
(2012). “Local tetra patterns: a new feature descriptor
for content-based image retrieval”, IEEE Transactions
on Image Processing, 21(5), 2874-2886.
[30]. Murala, S., Maheshwari, R. P., & Balasubramanian, R.
(2012). “Directional local extrema patterns: a new
descriptor for content based image
retrieval”, International journal of multimedia
information retrieval, 1(3), 191-203.
[31]. Murala, S., Maheshwari, R. P., & Balasubramanian, R.
(2012). “Directional binary wavelet patterns for
biomedical image indexing and retrieval”, Journal of
Medical Systems, 36(5), 2865-2879.
[32]. Murala, S., & Wu, Q. J. (2014). “Local mesh patterns
versus local binary patterns: biomedical image indexing
and retrieval’, IEEE Journal of Biomedical and Health
Informatics, 18(3), 929-938.
[33]. Murala, S., & Wu, Q. J. (2013). “Local ternary co-
occurrence patterns: a new feature descriptor for MRI
and CT image retrieval’, Neurocomputing, 119, 399-
412.
[34]. Reddy, P. V. B., & Reddy, A. R. M. (2014). “Content
based image indexing and retrieval using directional
local extrema and magnitude patterns”,AEU-
International Journal of Electronics and
Communications, 68(7), 637-643.
[35]. Jacob, I. J., Srinivasagan, K. G., & Jayapriya, K.
(2014). “Local oppugnant color texture pattern for
image retrieval system”, Pattern Recognition
Letters, 42, 72-78.
[36]. Corel 1000 and Corel 10000 image database. [Online].
Available: http://wang.ist.psu.edu/docs/related.shtml.
[37]. MIT Vision and Modeling Group, Vision Texture.
[Online]. Available: http://vismod.www.media.mit.edu.
[38]. Murala, S., & Wu, Q. J. (2014). “Local mesh patterns
versus local binary patterns: biomedical image indexing
and retrieval”, IEEE Journal of Biomedical and Health
Informatics, 18(3), 929-938.
[39]. H Kaur V Dhir (2016) “Local Maximum Edge co-
occurrence pattern for image retrieval”, IEEE Explore .