Docstoc

Performance Comparison of Texture Pattern Based Image Retrieval Methods using Walsh, Haar and Kekre Transforms with Assorted Thresholding Methods

Document Sample
Performance Comparison of Texture Pattern Based Image Retrieval Methods using Walsh, Haar and Kekre Transforms with Assorted Thresholding Methods Powered By Docstoc
					                                                            (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                    Vol. 09, No.03, 2011

   Performance Comparison of Texture Pattern Based
    Image Retrieval Methods using Walsh, Haar and
     Kekre Transforms with Assorted Thresholding
                      Methods
                                   Dr. H.B.Kekre1, Sudeep D. Thepade2, Varun K. Banura3
                     1
                      Senior Professor, 2Ph.D.Research Scholar & Associate Professor, 3B.Tech (CE) Student
                                          Computer Engineering Department, MPSTME,
                                  SVKM‟s NMIMS (Deemed-to-be University), Mumbai, India
                         1
                           hbkekre@yahoo.com, 2sudeepthepade@gmail.com,3varunkbanura@gmail.com

Abstract— Novel texture pattern based image retrieval                   could be listed as art galleries [15,17], museums, archaeology
techniques using image maps and non-sinusoidal orthogonal               [6], architecture design [11,16], geographic information
image transforms is the theme of the work presented in this             systems [8], weather forecast [8,25], medical imaging [8,21],
section. Different texture patterns namely ‘4-pattern’, ‘16-            trademark databases [24,26], criminal investigations [27,28],
pattern’, ‘64-pattern’ are generated using Haar transform               image search on the Internet [12,22,23]. The paper attempts to
matrix, Walsh transform matrix and Kekre transform matrix.              provide better and faster image retrieval techniques.
The generated texture patterns are then compared with the
image maps (binary image maps for Walsh patterns and Ternary            A. Content Based Image Retrieval
image maps for Haar patterns & Kekre patterns) of an image to
generate the feature vector based on structural matching (as the            For the first time Kato et.al. [7] described the experiments
matching number of ones, minus ones per Walsh texture pattern           of automatic retrieval of images from a database by colour and
and matching number of ones, zeros, minus ones per Haar/Kekre           shape feature using the terminology content based image
texture pattern). Further the image maps are created using four         retrieval (CBIR). The typical CBIR system performs two major
thresholding methods as global, local, intermediate with 4 tiles        tasks [19,20] as feature extraction (FE), where a set of features
(intermediate-4) and intermediate with 9 tiles (intermediate-9).        called feature vector is generated to accurately represent the
Here total 36 variations of proposed novel image retrieval              content of each image in the database and similarity
methods using texture patterns are considered with three image          measurement (SM), where a distance between the query image
transforms (Walsh, Haar & Kekre), three variations in number            and each image in the database using their feature vectors is
of texture patterns (4, 16 & 64) and four different ways to             used to retrieve the top “closest” images [19,20,29].
generate image maps (with global, local, intermediate-4,
intermediate-9 thresholding methods). The proposed texture                   For feature extraction in CBIR there are mainly two
content based image retrieval (CBIR) techniques are tested on           approaches [8] feature extraction in spatial domain and feature
the image database with help of 55 queries (randomly selected 5         extraction in transform domain. The feature extraction in
from each of 11 image categories) fired on image database. The          spatial domain includes the CBIR techniques based on
performance comparison of texture pattern based CBIR methods            histograms [8], BTC [4,5,19], VQ [24,28,29]. The transform
is done with help of precision-recall crossover points.                 domain methods are widely used in image compression, as they
                                                                        give high energy compaction in transformed image [20,27]. So
   Keywords- CBIR; Walsh, Haar & Kekre Transforms;                      it is obvious to use images in transformed domain for feature
Texture; Patterns; Image Maps                                           extraction in CBIR [26]. But taking transform of image is time
                                                                        consuming. Reducing the size of feature vector using pure
                         I.   INTRODUCTION                              image pixel data in spatial domain and getting the improvement
    Today the information technology experts are facing                 in performance of image retrieval is shown in [1,2,3]. But the
technical challenges to store/transmit and index/manage image           problem of feature vector size still being dependent on image
data effectively to make easy access to the image collections of        size persists in [1,2,3]. Here the query execution time is further
tremendous size being generated due to large numbers of                 reduced by decreasing the feature vector size further and
images generated from a variety of sources (digital camera,             making it independent of image size. Many current CBIR
digital video, scanner, the internet etc.). The storage and             systems use the Euclidean distance [4-6,11-17] on the extracted
transmission is taken care of by image compression [4,7,8].             feature set as a similarity measure. The Direct Euclidian
The image indexing is studied in the perspective of image               Distance between image P and query image Q can be given as
database [5,9,10,13,14] as one of the promising and important           equation 1, where Vpi and Vqi are the feature vectors of image
research area for researchers from disciplines like computer            P and Query image Q respectively with size „n‟.
vision, image processing and database areas. The hunger of
superior and quicker image retrieval techniques is increasing
day by day. The significant applications for CBIR technology



                                                                   76                               http://sites.google.com/site/ijcsis/
                                                                                                    ISSN 1947-5500
                                                               (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                       Vol. 09, No.03, 2011
                                 n
                     ED         (Vpi  Vqi )   2
                                                         (1)                                1, if .B(i, j )  TB
                                                                                           
                                i 1
                                                                              BMb(i, j )                                                  (8)
                                                                                            1, if .B(i, j )  TB
                                                                                           
               II.    GENERATION OF IMAGE MAPS
    Image bitmaps are prepared using four different types of
thresholding considerations as global, local, intermediate-4
and intermediate-9. For global thresholding, the image maps             B. Generation of Global Ternary Image Maps
of colour image are generated using three independent red (R),              The ternary image maps are used for comparison with
green (G) and blue (B) components of image to calculate three           texture patterns generated using Haar and Kekre transforms as
individual colour thresholds and one overall luminance                  the patterns contain three values one, zero and minus-one.
threshold [35]. Let X={R(i,j),G(i,j),B(i,j)} where i=1,2,….m            Here first for each component (R, G, and B), the individual
and j=1,2,….,n; be an m×n color image in RGB space. Let the             colour threshold intervals (lower-Tshl, and higher-Tshh) are
individual colour thresholds be TR, TG and TB which could               calculated as shown in equations 9, 10 and 11.
be computed as per the equations given below as 2, 3 and 4.
Let the luminance threshold T be as given by equation 5.
                                                                             Tshrl  TR  TR  T    , Tshrh  TR  TR  T                   (9)
                            m     n
                   1
          TR            R(i, j)
                  m * n i 1 j 1
                                                        (2)
                                                                             Tshgl  TG  TG  T    , Tshgh  TG  TG  T                 (10)

                   1 m n
          TG            G(i, j)
                  m * n i 1 j 1
                                                        (3)                  Tshbl  TB  TB  T    , Tshbh  TB  TB  T                 (11)

                                                                           Then the individual color plane global ternary image maps
                1 m n
          TB         B(i, j)
               m * n i 1 j 1
                                                        (4)
                                                                        are computed (TMr, TMg and TMb) as given in equations 12,
                                                                        13 and 14. If a pixel value of respective color component is
                                                                        greater than the respective higher threshold interval (Tshh),
                                                                        the corresponding pixel position of the image map gets a value
                     TR  TG  TB                                       „one‟; else if the pixel value is lesser than the respective lower
              T                                        (5)
                                                                        threshold interval (Tshl), the corresponding pixel position of
                           3
                                                                        the image map gets a value of „minus one‟; otherwise it gets a
                                                                        value „zero‟.
A. Generation of Global Binary Image Maps
                                                                                           1,        if .R(i, j )  Tshrh
    In binary image maps using global thresholding, the image                             
                                                                            TMr (i, j )   0, if .Tshrl  R(i, j )  Tshrh
is converted in to ones and minus ones only. Binary image                                  1,                                           (12)
                                                                                                     if .R(i, j )  Tshrl
bitmaps are generated using the individual color component
threshold values (TR, TG, TB) as BMr, BMg and BMb. If a
pixel in each component (R, G, and B) is greater than or equal
to the respective threshold, the corresponding pixel position of                           1,        if .G(i, j )  Tshgh
                                                                                          
the bitmap will have a value „one‟ otherwise it will have a                 TMg (i, j )   0, if .Tshgl  G(i, j )  Tshgh
value „minus one‟, as given by equations 6, 7 and 8. The                                   1,       if .G(i, j )  Tshgl                (13)
                                                                                          
binary image maps are used for comparison with texture
patterns generated using Walsh transform.

                                                                                          1,        if .B(i, j )  Tshbh
                    1, if .R(i, j )  TR                                               
                                                                           TMb(i, j )   0, if .Tshbl  B(i, j )  Tshbh
      BMr(i, j )                                      (6)                               1,                                            (14)
                                                                                                    if .B(i, j )  Tshbl
                    1, if .R(i, j )  TR
                   


                   1, if .G(i, j )  TG                               C. Generation of Global Ternary Image Maps
                  
     BMg(i, j )                                       (7)                The binary or ternary image maps are generated using four
                   1, if .G (i, j )  TG
                                                                       sundry methods of thresholding as global, intermediate-4,
                                                                        intermediate-9 and local. For global thresholding based image



                                                                   77                                    http://sites.google.com/site/ijcsis/
                                                                                                         ISSN 1947-5500
                                                              (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                      Vol. 09, No.03, 2011
maps, the global threshold values are computed as average of              (4-pattern). The 4x4 Walsh transform matrix is given in 2(d)
all pixel intensity values in the respective plane of considered          and visualization of 16 Walsh transform patterns generated
colour image (as given by equations 2,3,4 and 5). In case of              using it is shown in 2(e), where black color represent the
intermediate-4 thresholding, the image is divided into four               values „1‟ in the pattern and values „-1‟ are represented by
equal non overlapping parts (as shown in (b) of figure 1) and             white color. The obtained Walsh texture patterns then are
the image map of each of the part is generated using average              resized as the size of image for which texture features have to
of pixels of only that part as threshold. In case of intermediate-        be extracted.
9 thresholding based image maps, the image is divided into
nine non overlapping equal parts. For local thresholding, each
non distinct 2x2 pixel window is considered separately. Figure
1 shows the pixel group consideration for respective
thresholding methods, where the gray shading indicates the
group of pixel values to be considered for threshold
calculation in respective thresholding methods and number of
such pixel groups possible are given by black lines.                       (a) 2x2 Walsh         (b) 2x2 Walsh            (c) Generated 4
                                                                               Matrix                Matrix                Walsh Texture
                                                                                                                         Patterns (4-pattern)




           (a) Global                   (b) Intermediate-4


                                                                           (d) 4x4 Walsh       (e) Generated 16 Walsh Texture Patterns
                                                                               Matrix                        (16-pattern)

                                                                                   Figure 2. Walsh Texture Pattern Generation
                                                                          B. Haar Texture Pattern Generation
       (c) Intermediate-9                    (d) Local                       The 2x2, 4x4 and 8x8 Haar transform matrices are used to
Figure 1. Pixel group consideration for respective thresholding           generate the 4, 16 and 64 Haar texture patterns respectively.
 methods considered in image map generation for CBIR with                 The generation of four and sixteen Haar texture patterns [32]
                        texture patterns                                  is shown in figure 3. 2x2 Haar transform matrix [9,30,31] is
                                                                          shown as 3(a), each row of this matrix is considered one at a
             III.   TEXTURE PATTERN GENERATION                            time and is multiplied with all rows of the same matrix to
    Using the non-sinusoidal transform matrices assorted                  generate Haar texture patterns as shown in 3(b). Figure 3(c)
texture patterns namely 4-pattern, 16-pattern and 64-pattern              gives the visualization 4 Haar texture patterns (4-pattern). The
are generated. To generate N2 texture patterns (N2-pattern)               4x4 Haar transform matrix is given in 3(d) and 16 Haar
texture patterns, NxN transform matrix is considered and the              transform patterns generated using it, are shown in 3(e), where
element wise multiplication of each row of the transform                  black colour represent the values „1‟ in the pattern, grey colour
matrix is taken with all possible rows of the same matrix                 represents values „0‟ and values „-1‟ are represented by white
(consideration of one row at a time gives one pattern). The               colour. The obtained Haar texture patterns then are resized as
texture patterns obtained are orthogonal in nature. The                   the size of image for which texture features have to be
generation methods of Walsh transform, Haar transform and                 extracted. All the generated texture patterns are orthogonal to
Kekre transform patterns are elaborated respectively in                   each other.
sections A, B and C as given below.
A. Walsh Texture Pattern Generation
   The 4, 16 and 64 Walsh texture patterns are generated using
Walsh transform matrices [21,22,26,36] of size 2x2, 4x4 and
8x8 respectively. The generated four and sixteen Walsh
texture patterns [34] are shown in figure 2, 2x2 Walsh
transform matrix is shown as 2(a), each row of this matrix is               (a) 2x2 Haar          (b) 2x2 Haar            (c) Generated 4
considered one at a time and is multiplied with all rows of the                Matrix                Matrix                 Haar Texture
same matrix to generate Walsh texture patterns as shown in                                                               Patterns (4-pattern)
2(b). Figure 2(c) gives the envisioned 4 Walsh texture patterns



                                                                     78                              http://sites.google.com/site/ijcsis/
                                                                                                     ISSN 1947-5500
                                                            (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                    Vol. 09, No.03, 2011
                                                                                      IV.     CBIR USING TEXTURE PATTERNS
                                                                            In all total thirty six variations of proposed CBIR method
                                                                        are possible using the four methods of image maps (alias local,
                                                                        global, intermediate-4 and intermediate-9), three image
                                                                        transforms (Haar, Kekre and Walsh) and three different sets of
                                                                        texture patterns (4-pattern, 16-pattern and 64-pattern). For
                                                                        feature extraction in CBIR using texture patterns first the
                                                                        image map is generated using appropriate thresholding method
  (d) 4x4 Haar        (e) Generated 16 Haar Texture Patterns
                                                                        (local or global or intermediate-4 or intermediate-9). Then the
     Matrix                        (16-pattern)
                                                                        desired texture pattern set is generated (4-pattern or 16-pattern
                                                                        or 64-pattern) using the corresponding image transform (Haar
          Figure 3. Haar Texture Pattern Generation
                                                                        or Kekre or Walsh).
C. Kekre Texture Pattern Generation
                                                                            To generate feature vectors the binary image map of the
   The 4, 16 and 64 Kekre texture patterns are generated using          image is compared with each pattern of the generated Walsh
Kekre transform matrices [33,36] of size 2x2, 4x4 and 8x8               texture patterns to find matching number of ones & minus
respectively. Figure 4 gives generation of four and sixteen             ones in case of CBIR with Walsh texture patterns. The feature
Kekre texture patterns. 2x2 Kekre transform matrix is shown             vactor will have two values (number of matching „1‟ & „-1‟)
as 4 (a), each row of this matrix is considered one at a time           per pattern per colour plane in Walsh texture patterns. The per
and is multiplied with all rows of the same matrix to generate          image feature vector size for Walsh texture pattern based
Kekre texture patterns as shown in 4 (b) with all the negative          CBIR is given by equation 15.
values are replaced by „-1‟. Figure 4 (c) gives visualization of
the 4 Kekre texture patterns (4-pattern). The 4x4 Kekre
transform matrix is given in 5.16 (d) and visualization of 16            Feature vector size=2*3*(no. of considered texture-pattern)             (15)
Kekre transform patterns generated using it is shown in 5.16
(e), where black colour represent the values „1‟ in the pattern,            In case of Haar or Kekre texture patterns based CBIR, he
grey colour represent the values „0‟ and values „-1‟ are                ternary image map of the image is compared with each pattern
represented by white colour. The obtained Kekre texture                 of Haar or Kekre texture patterns to find three values per
patterns then are resized as the size of image for which texture        colour plane per pattern as number of matching ones, zeros &
features have to be extracted.                                          minus ones. The feature vector is formed using all these
                                                                        number of matches (for „1‟, „0‟ and „-1‟). The size of the
                                                                        feature vector of the image for Haar or Kekre texture patterns
                                                                        based CBIR is given by equation 16. Table 1 shows the feature
                                                                        vector size for 4, 16 and 64 texture patterns of respective
                                                                        image transforms.

                                                                         Feature vector size=3*3*(no. of considered texture-pattern)             (16)
 (a) 2x2 Kekre         (b) 2x2 Kekre        (c) Generated 4
     Matrix                Matrix            Kekre Texture
                                           Patterns (4-pattern)               Table 1. Feature vector of image retrieval using texture paterns
                                                                                                                          16-              64-
                                                                          CBIR Technique              4-Pattern
                                                                                                                        Pattern          Pattern
                                                                            Walsh Texture
                                                                                                           8               32              128
                                                                              Patterns
                                                                            Haar Texture
                                                                                                           12              48              192
                                                                              Patterns
                                                                            Kekre Texture
                                                                                                           12              48              192
                                                                              Patterns
 (d) 4x4 Kekre       (e) Generated 16 Kekre Texture Patterns
     Matrix                        (16-pattern)
                                                                            Using three assorted texture pattern set generated using
         Figure 4. Kekre Texture Pattern Generation                     Walsh, Haar and Kekre transform matrices along with image
                                                                        maps formed by different thresholds namely global,
                                                                        intermediate-4, intermediate-9 and local, total 36 novel feature
                                                                        vector generation methods have been tested resulting into six
                                                                        new image retrieval techniques. The main advantage of
                                                                        proposed CBIR methods is reduced time complexity for query
                                                                        execution due to reduced size of feature vector resulting into



                                                                   79                                   http://sites.google.com/site/ijcsis/
                                                                                                        ISSN 1947-5500
                                                                              (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                      Vol. 09, No.03, 2011
faster image retrieval with better performance. Also the                                pattern based CBIR methods with respective image map
feature vector size is independent of image size in proposed                            thresholding techniques are shown in figure 6. In CBIR using
CBIR methods.                                                                           texture patterns intermediate-4 thresholding has given better
                                                                                        performance than other considered thresholding methods. For
                           V.     IMPLEMENTATION                                        each thresholding method 16 pattern has given better
    The implementation of the discussed CBIR techniques is                              performance than 4 or 64. Except global thresholding, for 16
done in MATLAB 7.0 using a computer with Intel Core 2 Duo                               patterns Haar transform based 16-patterns have given better
Processor T8100 (2.1GHz) and 2 GB RAM. The CBIR                                         performance. Except local thresholding, for 64 pattern Kekre
techniques are tested on the Wang image database [18] of                                transform has shown better performance.
1000 variable size images spread across 11 categories of
human being, animals, natural scenery and manmade things,
etc. The categories and distribution of the images is shown in
table 2.
              Table 2. Image Database: Category-wise Distribution

  Category               Tribes              Buses              Beaches
    No. of
                            85                 99                     99
   Images
  Category               Horses           Mountains            Airplanes
    No. of
                            99                 61                     100
   Images
  Category             Dinosaurs          Elephants              Roses
    No. of                                                                               Figure 5. Crossover points of precision-recall for proposed texture pattern
                            99                 99                     99
   Images                                                                                  based CBIR methods with respect to the considered image transforms
  Category           Monuments              Sunrise
    No. of
                            99                 61
   Images

To assess the retrieval effectiveness, we have used the
precision and recall as statistical comparison parameters [4,5]
for the proposed CBIR techniques. The standard definitions
for these two measures are given by the equations 17 and 18.
                 Number _ of _ relevant _ images _ retrieved
Pr ecision                                                                 (17)
                  Total _ number _ of _ images _ retrieved
                  Number _ of _ relevant _ images _ retrieved
Re call                                                                    (18)
            Total _ number _ of _ relevent _ images _ in _ database

                                                                                         Figure 6. Crossover points of precision-recall for proposed texture pattern
                                                                                         based CBIR methods with respective image map thresholding techniques
        VI.       RESULTS OF CBIR USING TEXTURE PATTERNS
                                                                                        Figure 7 gives crossover points of precision-recall for
    The crossover point of precision-recall plays very                                  proposed texture pattern based CBIR methods with
important role in performance comparison of image retrieval                             corresponding number of patterns considered. Here 16 texture
methods, higher crossover point value indicates better image                            patterns have shown better performance than 4 or 64 texture
retrieval. The crossover points of average precision–recall                             patterns. In case of 4-texture patterns all transforms have
values of firing 55 queries on image database for proposed                              shown same performance (because of the 2x2 transform
texture pattern based image retrieval methods are computed                              matrices for all transforms are alike). In 16 patterns except
and plotted in figures 5, 6 and 7. Figure 5 gives crossover                             global thresholding Haar transform performs better. In case of
points of precision-recall for proposed texture pattern based                           64 patterns better performance is given by Kekre transform
CBIR methods with respect to the considered image                                       except local thresholding. The Haar 16 pattern based CBIR
transforms alias Walsh, Haar and Kekre. In all transform                                with intermediate 4 thresholding has shown best performance
texture pattern based CBIR methods except Walsh local                                   among all CBIR variations considered. In texture pattern
thresholding, 16 patterns have consistently performed well.                             based CBIR, image retrieval using the Haar 16 patterns with
Also for all transforms intermediate-4 thresholding has given                           intermediate 4 thresholding has given best performance with
better performance in all 4, 16 and 64 texture patterns. The                            precision-recall crossover point value 0.461524. The second
crossover points of precision-recall for proposed texture                               best performance with precision-recall crossover point value



                                                                                   80                                   http://sites.google.com/site/ijcsis/
                                                                                                                        ISSN 1947-5500
                                                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                  Vol. 09, No.03, 2011
0.45 is given by CBIR with Kekre 16 patterns with                                   Table 4. Performance Comparison of Number of Texture Patterns considered
                                                                                                     for image retrieval using Texture Patterns
intermediate 4 thresholding, the next in the performance is
image retrieval based on Kekre 16 patterns with global                                                            Number of             Average
                                                                                            Comparative
thresholding with crossover point value 0.4489 followed by                                                         Texture             Crossover
                                                                                            Performance
Haar 16 patterns with global thresholding and crossover point                                                      Patterns           Point Value
value 0.44834.                                                                                  Best              16 Pattern            0.43466
                                                                                             Second Best          64 Pattern           0.415209
                                                                                               Poorest             4 Pattern           0.412495

                                                                                     Table. 5 Performance Comparison of Thresholding method used to prepare
                                                                                               image maps for image retrieval using Texture Patterns

                                                                                                                                          Average
                                                                                            Comparative          Thresholding
                                                                                                                                         Crossover
                                                                                            Performance            Method
                                                                                                                                        Point Value
                                                                                               Best             Intermediate 4            0.43597
                                                                                            Second Best              Global               0.42180
                                                                                             Third Best          Intermediate 9           0.41363
 Figure 7. Crossover points of precision-recall for proposed texture pattern                  Poorest                 Local               0.41171
  based CBIR methods with corresponding number of patterns considered

 VII. PERFORMANCE COMPARISION OF VARIANTS IN TEXTURE                                   Total four varied thresholding methods are considered for
                  PATTERN BASED CBIR METHODS                                        image maps preparation for image retrieval using texture
                                                                                    patterns whose performance comparison is given in table 5 by
   The novel image retrieval methods using texture patterns
                                                                                    means of average precision-recall crossover point values of
are presented in this section. Here in all 36 variations of
                                                                                    texture based CBIR variants using respective thresholding
proposed image retrieval methods with texture patterns are
                                                                                    method. Intermediate 4 thresholding has been proven better.
proposed using three image transforms (Haar, kekre & Walsh),
                                                                                    The performance ranking for thresholding methods used in
three types of texture patterns (16, 64 & 4) and four ways of
                                                                                    proposed CBIR with texture patterns, starting with the best can
thresholding used to prepare image maps (intermediate-4,
                                                                                    be given as intermediate 4, global, intermediate 9 and local.
global, intermediate-9 & local). The average of precision-
recall crossover point values for respective variation is                                                     VIII. REFERENCES
considered for the performance ranking of these variations.                         [1]   Dr. H.B.Kekre, Sudeep D. Thepade, Varun K. Banura, “Augmentation
Three image transforms namely Walsh, Haar and Kekre are                                   of Colour Averaging Based Image Retrieval Techniques using Even
considered to generate texture patterns. From the results after                           part of Images and Amalgamation of feature vectors”, International
experimentation it is found that the Haar transform is showing                            Journal of Engineering Science and Technology (IJEST), Volume 2,
                                                                                          Issue 10, (ISSN: 0975-5462) Available online at http://www.ijest.info
best performance followed by Kekre transform and then
                                                                                    [2]   Dr. H.B.Kekre, Sudeep D. Thepade, Varun K. Banura, “Amelioration
Walsh transform in proposed CBIR methods as indicated by                                  of Colour Averaging Based Image Retrieval Techniques using Even
average precision-recall crossover point values of texture                                and Odd parts of Images”, International Journal of Engineering Science
based CBIR variants using respective image transform given                                and Technology (IJEST), Volume 2, Issue 9, (ISSN: 0975-5462)
in table 3. The number of texture patterns considered here are                            Available online at http://www.ijest.info.
4, 16 and 64. The 16 texture patterns have shown better                             [3]   Dr. H.B.Kekre, Sudeep D. Thepade, Akshay Maloo, “Query by Image
                                                                                          Content Using Colour Averaging Techniques”, International Journal of
performance in CBIR using texture patterns. The 64 texture                                Engineering Science and Technology (IJEST), Volume 2, Issue 6,
patterns have given second best performance followed by 4                                 2010.pp.1612-1622 (ISSN: 0975-5462) Available online at
texture patterns with poorest performance as per the average                              http://www.ijest.info.
precision-recall crossover point values of texture based CBIR                       [4]   Dr. H.B.Kekre, Sudeep D. Thepade, “Boosting Block Truncation
variants using respective number of texture patterns given in                             Coding using Kekre‟s LUV Color Space for Image Retrieval”, WASET
                                                                                          International Journal of Electrical, Computer and System Engineering
table 4.                                                                                  (IJECSE), Volume 2, Number 3, pp. 172-180, Summer 2008. Available
   Table 3. Performance Comparison of Image Transforms used in image                      online at http://www.waset.org/ijecse/v2/v2-3-23.pdf
                     retrieval using Texture Patterns                               [5]   Dr. H.B.Kekre, Sudeep D. Thepade, “Image Retrieval using
                                                                                          Augmented Block Truncation Coding Techniques”, ACM International
                                                      Average                             Conference on Advances in Computing, Communication and Control
        Comparative              Image
                                                     Crossover                            (ICAC3-2009), pp. 384-390, 23-24 Jan 2009, Fr. Conceicao Rodrigous
        Performance            Transform
                                                    Point Value                           College of Engg., Mumbai. Is uploaded on online ACM portal.
           Best                   Haar                0.42320                       [6]   Dr. H.B.Kekre, Sudeep D. Thepade, “Scaling Invariant Fusion
        Second Best               Kekre               0.41998                             of Image Pieces in Panorama Making and Novel Image
                                                                                          Blending Technique”, International Journal on Imaging (IJI),
          Poorest                 Walsh               0.41916                             www.ceser.res.in/iji.html, Volume 1, No. A08, pp. 31-46, Autumn
                                                                                          2008.




                                                                               81                                   http://sites.google.com/site/ijcsis/
                                                                                                                    ISSN 1947-5500
                                                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                  Vol. 09, No.03, 2011
[7]    Hirata K. and Kato T. “Query by visual example – content-based image            [23]   Dr. H.B.Kekre, Sudeep D. Thepade, “Color Based Image Retrieval
       retrieval”, In Proc. of Third International Conference on Extending                    using Amendment Block Truncation Coding with YCbCr Color Space”,
       Database Technology, EDBT‟92, 1992, pp 56-71                                           International Journal on Imaging (IJI), Volume 2, Number A09,
[8]    Dr. H.B.Kekre, Sudeep D. Thepade, “Rendering Futuristic Image                          Autumn 2009, pp. 2-14. Available online at www.ceser.res.in/iji.html.
       Retrieval System”, National Conference on Enhancements in Computer,             [24]   Dr. H.B.Kekre, Tanuja Sarode, Sudeep D. Thepade, “Color-Texture
       Communication and Information Technology, EC2IT-2009, 20-21 Mar                        Feature based Image Retrieval using DCT applied on Kekre‟s Median
       2009, K.J.Somaiya College of Engineering, Vidyavihar, Mumbai-77.                       Codebook”, International Journal on Imaging (IJI), Volume 2, Number
[9]    Minh N. Do, Martin Vetterli, “Wavelet-Based Texture Retrieval Using                    A09,      Autumn      2009,pp.    55-65.    Available     online      at
       Generalized Gaussian Density and Kullback-Leibler Distance”, IEEE                      www.ceser.res.in/iji.html (ISSN: 0974-0627).
       Transactions On Image Processing, Volume 11, Number 2, pp.146-158,              [25]   Dr. H.B.Kekre, Sudeep D. Thepade, Akshay Maloo “Performance
       February 2002.                                                                         Comparison for Face Recognition using PCA, DCT &WalshTransform
[10]   B.G.Prasad, K.K. Biswas, and S. K. Gupta, “Region –based image                         of Row Mean and Column Mean”, ICGST International Journal on
       retrieval using integrated color, shape, and location index”,                          Graphics, Vision and Image Processing (GVIP), Volume 10, Issue II,
       International Journal on Computer Vision and Image Understanding                       Jun.2010,              pp.9-18,           Available              online
       Special Issue: Colour for Image Indexing and Retrieval, Volume 94,                     http://209.61.248.177/gvip/Volume10/Issue2/P1181012028.pdf..
       Issues 1-3, April-June 2004, pp.193-233.                                        [26]   Dr. H.B.Kekre, Sudeep D. Thepade, “Improving the Performance of
[11]   Dr. H.B.Kekre, Sudeep D. Thepade, “Creating the Color Panoramic                        Image Retrieval using Partial Coefficients of Transformed Image”,
       View using Medley of Grayscale and Color Partial Images ”, WASET                       International Journal of Information Retrieval, Serials Publications,
       International Journal of Electrical, Computer and System Engineering                   Volume 2, Issue 1, 2009, pp. 72-79 (ISSN: 0974-6285)
       (IJECSE), Volume 2, No. 3, Summer 2008. Available online at                     [27]   Dr. H.B.Kekre, Sudeep D. Thepade, Archana Athawale, Anant Shah,
       www.waset.org/ijecse/v2/v2-3-26.pdf.                                                   Prathmesh Verlekar, Suraj Shirke, “Performance Evaluation of Image
[12]   Stian Edvardsen, “Classification of Images using color, CBIR Distance                  Retrieval using Energy Compaction and Image Tiling over DCT Row
       Measures and Genetic Programming”, Ph.D. Thesis, Master of science                     Mean and DCT Column Mean”, Springer-International Conference on
       in Informatics, Norwegian university of science and Technology,                        Contours of Computing Technology (Thinkquest-2010), Babasaheb
       Department of computer and Information science, June 2006.                             Gawde Institute of Technology, Mumbai, 13-14 March 2010, The paper
                                                                                              will be uploaded on online Springerlink.
[13]   Dr. H.B.Kekre, Tanuja Sarode, Sudeep D. Thepade, “DCT Applied to
       Row Mean and Column Vectors in Fingerprint Identification”, In                  [28]   Dr. H.B.Kekre, Tanuja K. Sarode, Sudeep D. Thepade, Vaishali
       Proceedings of International Conference on Computer Networks and                       Suryavanshi,“Improved Texture Feature Based Image Retrieval using
       Security (ICCNS), 27-28 Sept. 2008, VIT, Pune.                                         Kekre‟s Fast Codebook Generation Algorithm”, Springer-International
                                                                                              Conference on Contours of Computing Technology (Thinkquest-2010),
[14]   Zhibin Pan, Kotani K., Ohmi T., “Enhanced fast encoding method for
                                                                                              Babasaheb Gawde Institute of Technology, Mumbai, 13-14 March
       vector quantization by finding an optimally-ordered Walsh transform
                                                                                              2010, The paper will be uploaded on online Springerlink.
       kernel”, ICIP 2005, IEEE International Conference, Volume 1, pp I -
       573-6, Sept. 2005.                                                              [29]   Dr. H.B.Kekre, Tanuja K. Sarode, Sudeep D. Thepade, “Image
                                                                                              Retrieval by Kekre‟s Transform Applied on Each Row of Walsh
[15]   Dr. H.B.Kekre, Sudeep D. Thepade, “Improving „Color to Gray and
                                                                                              Transformed VQ Codebook”, (Invited), ACM-International Conference
       Back‟ using Kekre‟s LUV Color Space”, IEEE International Advanced
                                                                                              and Workshop on Emerging Trends in Technology (ICWET
       Computing Conference 2009 (IACC‟09), Thapar University, Patiala,
                                                                                              2010),Thakur College of Engg. And Tech., Mumbai, 26-27 Feb 2010,
       INDIA, 6-7 March 2009. Is uploaded at online at IEEE Xplore.
                                                                                              The paper is invited at ICWET 2010. Also will be uploaded on online
[16]   Dr. H.B.Kekre, Sudeep D. Thepade, “Image Blending in Vista Creation                    ACM Portal.
       using Kekre's LUV Color Space”, SPIT-IEEE Colloquium and
                                                                                       [30]   Haar, Alfred, “ZurTheorie der orthogonalenFunktionensysteme”.
       International Conference, Sardar Patel Institute of Technology, Andheri,
                                                                                              (German), MathematischeAnnalen, volume 69, No. 3, 1910, pp. 331–
       Mumbai, 04-05 Feb 2008.
                                                                                              371.
[17]   Dr. H.B.Kekre, Sudeep D. Thepade, “Color Traits Transfer to
                                                                                       [31]   Charles K. Chui, “An Introduction to Wavelets”, Academic Press, 1992,
       Grayscale Images”, In Proc.of IEEE First International Conference on
                                                                                              San Diego, ISBN 0585470901.
       Emerging Trends in Engg. & Technology, (ICETET-08), G.H.Raisoni
       COE, Nagpur, INDIA. Uploaded on online IEEE Xplore.                             [32]   Dr. H. B. Kekre, Sudeep D. Thepade, Varun K. Banura, “Query by
                                                                                              Image Texture Pattern content using Haar Transform Matrix and Image
[18]   http://wang.ist.psu.edu/docs/related/Image.orig (Last referred on 23
                                                                                              Bitmaps”, Invited at ACM International Conference and Workshop on
       Sept 2008)
                                                                                              Emerging Trends in Technology (ICWET 2011), TCET, Mumbai, 25-
[19]   Dr. H.B.Kekre, Sudeep D. Thepade, “Using YUV Color Space to Hoist                      26 Feb 2011.
       the Performance of Block Truncation Coding for Image Retrieval”,
                                                                                       [33]   Dr. H. B. Kekre, Sudeep D. Thepade, “Image Retrieval using Non-
       IEEE International Advanced Computing Conference 2009 (IACC‟09),
                                                                                              Involutional Orthogonal Kekre‟s Transform”, International Journal of
       Thapar University, Patiala, INDIA, 6-7 March 2009.
                                                                                              Multidisciplinary Research and Advances in Engineering (IJMRAE),
[20]   Dr. H.B.Kekre, Sudeep D. Thepade, Archana Athawale, Anant Shah,                        Ascent Publication House, 2009, Volume 1, No.I, pp 189-203, 2009.
       Prathmesh Verlekar, Suraj Shirke,“Energy Compaction and Image                          Abstract available online at www.ascent-journals.com
       Splitting for Image Retrieval using Kekre Transform over Row and
                                                                                       [34]   Dr. H. B. Kekre, Sudeep D. Thepade, Varun K. Banura, “Image
       Column Feature Vectors”, International Journal of Computer Science
                                                                                              Retrieval using Texture Patterns generated from Walsh-Hadamard
       and Network Security (IJCSNS),Volume:10, Number 1, January 2010,
                                                                                              Transform Matrix and Image Bitmaps”, (Selected) Springer
       (ISSN: 1738-7906) Available at www.IJCSNS.org.
                                                                                              International Conference on Technology Systems and Management
[21]   Dr. H.B.Kekre, Sudeep D. Thepade, Archana Athawale, Anant Shah,                        (ICTSM 2011), MPSTME and DJSCOE, Mumbai, 25-27 Feb 2011.
       Prathmesh Verlekar, Suraj Shirke, “Walsh Transform over Row Mean
                                                                                       [35]   Dr. H.B.Kekre, Sudeep D. Thepade, Shrikant Sanas, “Improving
       and Column Mean using Image Fragmentation and Energy Compaction
                                                                                              Performance of multileveled BTC based CBIR using Sundry Color
       for Image Retrieval”, International Journal on Computer Science and
                                                                                              Spaces”, CSC International Journal of Image Processing (IJIP), Volume
       Engineering (IJCSE),Volume 2S, Issue1, January 2010, (ISSN: 0975–
                                                                                              4, Issue 6, Computer Science Journals, CSC Press,
       3397). Available online at www.enggjournals.com/ijcse.
                                                                                              www.cscjournals.org
[22]   Dr. H.B.Kekre, Sudeep D. Thepade, “Image Retrieval using Color-
                                                                                       [36]   Dr. H.B.Kekre, Sudeep D. Thepade, Akshay Maloo “Performance
       Texture Features Extracted from Walshlet Pyramid”, ICGST
                                                                                              Comparison of Image Retrieval Using Fractional Coefficients of
       International Journal on Graphics, Vision and Image Processing
                                                                                              Transformed Image Using DCT, Walsh, Haar and Kekre‟s Transform”,
       (GVIP), Volume 10, Issue I, Feb.2010, pp.9-18, Available online
                                                                                              CSC International Journal of Image Processing (IJIP), Volume 4, Issue
       www.icgst.com/gvip/Volume10/Issue1/P1150938876.html




                                                                                  82                                    http://sites.google.com/site/ijcsis/
                                                                                                                        ISSN 1947-5500
                                                                (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                        Vol. 09, No.03, 2011
2, pp 142-157, Computer         Science    Journals,   CSC     Press,                  Management and Engineering, SVKM‟s NMIMS University,
www.cscjournals.org                                                                    Vile Parle(w),     Mumbai, INDIA. He is member of
                                                                                       International Association of Engineers (IAENG) and
                                                                                       International Association of Computer Science and
                 AUTHORS PROFILE
                                                                                       Information Technology (IACSIT), Singapore. He has been
                                                                                       on International Advisory Board of many International
       Dr. H. B. Kekre has received B.E. (Hons.) in Telecomm.                          Conferences. He is Reviewer for many reputed International
       Engineering. from Jabalpur University in 1958, M.Tech                           Journals. His areas of interest are Image Processing and
       (Industrial Electronics) from IIT Bombay in 1960,                               Computer Networks. He has more than 100 research papers
       M.S.Engg. (Electrical Engg.) from University of Ottawa in                       in National/International Conferences/Journals to his credit
       1965 and Ph.D. (System Identification) from IIT Bombay                          with a Best Paper Award at International Conference
       in 1970 He has worked as Faculty of Electrical Engg. and                        SSPCCIN-2008, Second Best Paper Award at ThinkQuest-
       then HOD Computer Science and Engg. at IIT Bombay. For                          2009 National Level paper presentation competition for
       13 years he was working as a professor and head in the                          faculty, Best Paper Award at Springer International
       Department of Computer Engg. at Thadomal Shahani                                Conference ICCCT-2010 and second best project award at
       Engineering. College, Mumbai. Now he is Senior Professor                        Manshodhan 2010.
       at MPSTME, SVKM‟s NMIMS University. He has guided
       17 Ph.Ds, more than 100 M.E./M.Tech and several
       B.E./B.Tech projects. His areas of interest are Digital Signal                  Varun K. Banura is currently pursuing B.Tech. (CE) from
       processing, Image Processing and Computer Networking. He                        MPSTME, NMIMS University, Mumbai. His areas of
       has more than 320 papers in National / International                            interest are Image Processing and Computer Networks. He
       Conferences and Journals to his credit. He was Senior                           has 07 research papers in International Conferences/Journals
       Member of IEEE. Presently He is Fellow of IETE and Life                         to his credit.
       Member of ISTE Recently ten students working under his
       guidance have received best paper awards and two have been
       conferred Ph.D. degree of SVKM‟s NMIMS University.
       Currently 10 research scholars are pursuing Ph.D. program
       under his guidance.


       Sudeep D. Thepade has Received B.E.(Computer) degree
       from North Maharashtra University with Distinction in 2003.
       M.E. in Computer Engineering from University of Mumbai
       in 2008 with Distinction, currently pursuing Ph.D. from
       SVKM‟s NMIMS, Mumbai. He has about than 07 years of
       experience in teaching and industry. He was Lecturer in
       Dept. of Information Technology at Thadomal Shahani
       Engineering College, Bandra(w), Mumbai for nearly 04
       years. Currently working as Associate Professor in Computer
       Engineering at Mukesh Patel School of Technology




                                                                        83                             http://sites.google.com/site/ijcsis/
                                                                                                       ISSN 1947-5500