Docstoc

Performance Appraise of Assorted Thresholding Methods in CBIR using Block Truncation Coding

Document Sample
Performance Appraise of Assorted Thresholding Methods in CBIR using Block Truncation Coding Powered By Docstoc
					                                                                (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                   Vol. 09, No.06, June 2011


        Performance Appraise of Assorted Thresholding
        Methods in CBIR using Block Truncation Coding
                                         Dr. H.B.Kekre1, Sudeep D. Thepade2, Shrikant Sanas3
                           1
                            Senior Professor, 2Associate Professor & Ph.D.Research Scholar, 3M.Tech Student
                                             Computer Engineering Department, MPSTME,
                                      SVKM‟s NMIMS (Deemed-to-be University), Mumbai, India
                              1
                                hbkekre@yahoo.com, 2sudeepthepade@gmail.com,3shri.sanas@gmail.com

Abstract— The paper proposes various types of thresholding                  to provide a high percentage of relevant images in response to
methods for generation of image bitmaps used in Block                       the query image. The goal of an image retrieval system is to
Truncation Coding (BTC), also the performance comparison                    retrieve a set of images from a collection of images such that this
of these assorted thresholding methods in image retrieval                   set meets the use requirements [14,15,16]. The user‟s
using multilevel BTC is presented. The different hresholding                requirements can be specified in terms of similarity to some
methods discussed here alias Global thresholding, Local                     other image.
thresholding and Intermediate thresholding. Based on the
type of thresholding method used for bitmap generation in                   The challenge to image indexing/management is studied in the
BTC the performance of the respective BTC based image                       context of image database, Image retrieval systems can be
retrieval method varies. The proposed variations of BTC                     divided into two main types: Text Based Image Retrieval and
based image retrieval techniques are tested on extended                     Content Based Image Retrieval [9].
Wang generic image database of 1000 images spread across
11 categories. For each CBIR (content based image retrieval)                Text Based Image Retrieval is the traditional image retrieval
technique, 1000 queries are fired on image database to                      system. In retrieval systems by adding text strings describing the
compute average precision and recall for all queries with                   content of an image. This system having drawbacks first It is
respect to number of retrieved image. These values are                      very time consuming technique. Second as per perception of
plotted to obtain the crossover point of precision and recall,              information from same image different person having different
which is used as criteria for performance comparison. The                   meaning. Due to these drawbacks, Content Based Image
results have shown the performance improvement (i.e.,                       Retrieval (CBIR) is introduced [20].
higher precision and recall crossover point values) with
Intermediate BTC-CBIR method. The performance of                            A Content Based Image Retrieval (CBIR) gives query as Image
multileveled Intermediate BTC-CBIR increases gradually                      and output is number of matching images to query image. In a
with increase in level up to certain extent (Level 3) and then              CBIR, features are used to represent the image content. The
increases slightly due to voids being created at higher levels.             features are extracted automatically and there is no manual
In all level 3 of BTC Intermediate-9 BTC gives best                         intervention, thus eliminating the dependency on humans in the
performance.                                                                feature extraction stage. The typical CBIR system performs two
                                                                            major tasks. The first one is feature extraction (FE), where a set
Keywords: CBIR, BTC, Multilevel BTC, Thresholding.                          of features, called feature vector, is generated to accurately
                                                                            represent the content of each image in the database. A feature
1. Introduction                                                             vector is much smaller in size than the original image. The
Visual communication plays an important role in human                       second task is similarity measurement (SM), where a distance
communication. We live in digital era with advancement in                   between the query image and each image in the database using
information and communication technology. Large amount of                   their signatures is computed so that the top “closest” images can
digital data is generated, transmitted, stored, analyzed and                be retrieved [3, 11, 12, 13]. Many current CBIR system use
accessed. Mostly information is in multimedia nature such as                Euclidean distance on the extracted feature set as a similarity
digital images, audio, video, graphics. From various sources                measure. The Direct Euclidian distance [21] between image P
large amount of images are generated and it takes large volume              and query image Q can be given as equation 1. Where Vpi and
for storage purpose [1,2,3,4]. This store information in the form           Vqi are the feature vectors of image P and query image Q
of images are more complex to retrieve and it is difficult to store         respectively with size „n‟.
in large volume. The need for efficient retrieval of images has
                                                                                                            n
been recognized by managers of large image collections [5,6,7].
To develop efficient indexing techniques for the retrieval of
                                                                                              ED  1 / 2    (Vpi  Vqi)2                      (1)
                                                                                                           i 1
enormous volumes of images being generated these days,
reasonable solutions need to be achieved. Content based image               Where, Vpi and Vqi be the feature vectors of image P and Query
retrieval (CBIR) is one of such attempts [8,9,10]. CBIR is used             image Q respectively with size „n‟.




                                                                      249                            http://sites.google.com/site/ijcsis/
                                                                                                     ISSN 1947-5500
                                                                         (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                            Vol. 09, No.06, June 2011

The thirst of better and faster image retrieval techniques is                  For intermediate-4 thresholding, the image is divided into four
increasing day by day. Some of important applications for CBIR                 quadrants (as shown in figure 1.c) and threshold for each
technology could be identified as art galleries [21], museums,                 quadrant is considered separately, to divide pixels of each colour
archaeology, architecture design, geographic information                       plane of the image into upper and lower clusters respectively;
systems, weather forecast, medical imaging, trademark                          resulting into feature vector of size six values (two per colour
databases, criminal investigations [19], image search on the                   plane, upper average and lower average) used in Intermediate-4
internet.                                                                      BTC based CBIR.

2. Block Truncation Coding (BTC)                                               In the similar manner if the image is divided into nine non
Block truncation coding (BTC) is a relatively simple and basic                 overlapping equal parts (as shown in figure 1.d) and threshold
image coding or compression technique [14]. Many advanced                      over each part is considered to divide the pixels in that part into
coding techniques were developed based on this technique. It                   upper and lower clusters, intermediate-9 thresholding is used.
was initially developed in 1979 for grayscale image coding. In                 The generated feature vector with upper mean and lower mean
this technique, the image to be compressed or coded is divided                 per colour plane of the image is used in Intermediate-9 BTC
into small non-overlapping blocks of size 4x4 pixels or of size                based CBIR.
which gives a reasonable resolution. Coding is done on one
block at a time. A binary bit-map is created for each pixel within
the block based on the block average mean. If the pixel value is
greater than the mean pixel value of the block, then value of 1 is
marked in the bitmap in the respective pixel position. Value of 0
is marked if the pixel value is less than the block mean value.

As a next step, the bitmap table is divided into two bitmap tables
called “upper mean” and “lower mean”. Upper mean bitmap
table is created for all pixel positions having 1 as value and
lower mean bitmap is created for all pixels having 0 as value.
The upper mean table will take the original pixel value for the
pixel position that has value 1 and 0 for rest of the pixel position.
Similarly, the lower mean bitmap table will take original pixel
value for the pixel positions that has 0 values and 0 for the rest
of pixel positions.

3. Thresholding Methods
In BTC first the average of the image (referred as „threshold‟) is
taken and then it is used to divide all image pixels into two
groups, greater than average and lesser than average. The means                    Figure 1. Proposed Thresholding Methods to be used in BTC
of each of these two groups per colour plane are considered as
feature vectors for BTC based CBIR. In conventional BTC the
threshold is computed by taking mean of all image pixels per                   4. CBIR using Global BTC level-1 (BTC-6) [3, 4, 5, 18, 22]
plane which here is referred as Global thesholding. These means                In original BTC the image is divided into R, B, and G
per colour plane when concatenated together form feature vector                components and the inter band average image (IBAI) is
of respective image to be used in Global BTC based CBIR. As                    computed, which is the average of all the components(R, G, and
shown in figure 1 there can be other possibilities of computing                B) and mean of inter band average image is taken as threshold.
the thresholds for dividing the image data into two clusters                   By using three independent R, G and B components of color
(greater and lesser than the threshold).                                       images to calculate three different thresholds and then apply
                                                                               BTC to each individual R, G and B planes. Let the thresholds be
The approach of local thresholding considers each 2x2 pixel                    MR, MG and MB, which could be computed as per the equations
block separately to compute the threshold (as shown in figure                  2, 3 and 4 given below.
1.b) and further all the pixels having value greater than local                                     1 m n
thresholds are grouped as upper cluster and other pixels having                            MR            R(i, j )
                                                                                                   m * n i 1 j 1
                                                                                                                                                (2)
values lower than the local thresholds form lower cluster,
resulting again into feature vector having six values with lower
average and upper average per plane, used for Local BTC based
                                                                                                         m     n
CBIR.                                                                                                                                           (3)
                                                                                                          G(i, j)
                                                                                                    1
                                                                               X            MG 
                                                                                                   m*n   i 1 j 1




                                                                        250                              http://sites.google.com/site/ijcsis/
                                                                                                         ISSN 1947-5500
                                                                                              (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                                 Vol. 09, No.06, June 2011

                                                                                                    5. CBIR using Multilevel Global BTC [1][2]
                                         m     n                                 (4)                In Multilevel BTC [1,2] the colour averages are found at
                                         B(i, j)
                                   1
V                  MB                                                                              different levels of Block Truncation Coding. The increase in the
                                  m*n    i 1 j 1
                                                                                                    level of BTC results into increased feature vector size. The
Then the three binary bitmaps are computed as BMr, BMg and                                          feature vector at a particular level of BTC is used for retrieving
BMb respectively as given in equations 5, 6 and 7. If a pixel in                                    images from large database. In the paper the multilevel BTC is
each component (R, G, and B) is greater than or equal to the                                        applied on local and intermediate thresholding methods.
respective threshold, the corresponding pixel position of the
bitmap will have a value of 1 otherwise it will have a value of 0.

                             1, if ....R(i, j )  MR
             BMr (i, j )  {                                                     (5)
                            0,....if ...R(i, j )  MR

                             1, if ....G (i, j )  MG
             BMg (i, j )  {                                                     (6)
                            0,....if ...G (i, j )  MG
                                                                                                                            Figure 2: Multilevel BTC
                                      1, if ....B(i, j )  MB
                                                                                                    CBIR using BTC-Level 2 (BTC-12)
            BMb(i, j )  {                                                       (7)                In BTC- Level 2 the image data is divided into 12 parts using the
                          0,....if ...B(i, j )  MB                                                six means obtained in BTC-Level 1. Here the bitmap are
                                                                                                    prepared using upper and lower mean values of individual colour
Two mean colors one for the pixels greater than or equal to the                                     components. For red color component the bitmap „BMUR‟ and
threshold and other for the pixels smaller than the threshold are                                   „BMLR‟ are generated as given in equations 14 and 15.
also calculated [15]. The upper mean color UM (UR, UG, UB) is                                       Similarly for green color component „BMUG‟ & „BMLR‟ and
given as follows in equations 8,9 and 10.                                                           for blue color components „BMUB‟ & „BMLB‟ can be
                                                   m       n
                                  1
           UR     m      n
                                              *  BMr (i, j ) * R(i, j )                           generated.
                                                                                      (8)                                      1, if ....R(i, j )  UR
                   BMr(i, j )                 i 1 j 1

                  i 1 j 1                                                                                   BMUR(i, j )  {                                           (14)
                                                   m       n                                                                  0,....if ...R(i, j )  UR
                                  1
           UG     m      n
                                              *  BMg (i, j ) * G (i, j )                                                     1, if ....R(i, j )  LR
                                                                                      (9)
                   BMg (i, j )                i 1 j 1
                                                                                                              BMLR(i, j )  {                                           (15)
                  i 1 j 1
                                                                                                                             0,....if ...R(i, j )  LR
                                                   m       n
                                  1
           UB    m       n
                                              *  BMb(i, j ) * B(i, j )
                                                                                      (10)          Using this bitmap the two mean colors per bitmap one for the
                   BMb(i, j )                 i 1 j 1
                                                                                                    pixels greater than or equal to the threshold and other for the
                  i 1 j 1
                                                                                                    pixels smaller than the threshold are calculated [15]. The upper
                                                                                                    mean color UM (UUR, ULR, UUG, ULG, UUB, ULB) are
And the Lower Mean LM= (LR, LG, LB) is computed as                                                  given as equations 16 and 17.
following equations 11, 12 and 13.                                                                                     1           m  n

                                               m       n                                               UUR  m n                 *   BMUR(i, j ) * Iur (i, j )
                          1
                                           * {1  BMr (i, j )} * R(i, j )
                                                                                                                                                                 (16)
    LR             m     n
           m * n   BMr (i, j )             i 1 j 1                        (11)                              BMUR(i, j ) i 1 j 1
                                                                                                                i 1 j 1
                   i 1 j 1


                                                                                                                                         m    n
                                                                                                                            1
                                                                                                        ULR                          *   BMLR(i, j ) * Ilr (i, j )
                                               m       n
                          1
 LG                                       * {1  BMg (i, j )} * G (i, j )                                    m     n                                                  (17)
                                                                                                                 BMLR(i, j )
                   m      n
                                                                                                                                        i 1 j 1
        m * n   BMg (i, j )                i 1 j 1                        (12)
                  i 1 j 1                                                                                     i 1 j 1


                                               m       n
    LB 
                          1
                                             * {1  BMb(i, j )} * B(i, j )
                                                                                                    And the first two components of Lower Mean LM= (LUR, LLR,
                      m       n
           m * n   BMb(i, j )              i 1 j 1                        (13)                 LUG, LLG, LUB, LLB) are computed using following equations
                    i 1 j 1                                                                       18 and 19.




                                                                                             251                                http://sites.google.com/site/ijcsis/
                                                                                                                                ISSN 1947-5500
                                                                           (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                              Vol. 09, No.06, June 2011


                         1      m    n                                           precision and recall as statistical comparison parameter for the
     LUR                     *  {1  BMUR(i, j )}* Iur(i, j )
             m     n                                               (18)          BTC-6, BTC-12, BTC-24 and BTC-48 techniques of CBIR on
              BMUR(i, j )    i 1 j 1
                                                                                 Global thresholding, Local thresholding and Intermediate
             i 1 j 1
                                m    n
                                                                                 thresholding types. The standard definitions of these two
                         1
     LLR                     *  {1  BMLR(i, j )}* Ilr(i, j )                 measures are given by following equations 20 and 21.
             m     n                                               (19)
              BMLR(i, j )    i 1 j 1

                                                                                                    Number _ of _ relevant _ images _ retrieved
             i 1 j 1
                                                                                    Pr ecesion                                                          (20)
                                                                                                     Total _ number _ of _ images _ retrieved
These Upper Mean and Lower Mean together will form a feature
                                                                                                     Number _ of _ relevant _ images _ retrieved
vector for BTC-12. For every image stored in the database these                    Re call 
                                                                                               Total _ number _ of _ relevent _ images _ in _ database   (21)
feature vectors are computed and stored in feature vector table.

CBIR using BTC-Level 3 (BTC-24)                                                  8. Results and Discussion
Similarly the feature vector for BTC-24 can be found by                          The proposed BTC-CBIR methods are applied on image
extending the BTC till level 3 as shown in figure 1. Each plane                  database of 1000 images at four different levels of BTC such as
will give the 6 elements of feature vector. For Red plane we get                 BTC Level-1, BTC Level-2, BTC Level-3 and BTC Level-4. For
(UUUR, LUUR, ULUR, LLUR, UULR, LULR, ULLR, and                                   each level of BTC the four assorted thresholding methods as
LLLR).                                                                           global, local, intermediate-4 and intermediate-9 are considered
                                                                                 to get 16 variations of BTC based CBIR methods. The average
CBIR using BTC-Level 3[1] and BTC-RGB-Level 4 [2]                                precision and recall values of 1000 queries per CBIR method are
Similarly the feature vector for BTC-24 can be found by                          computed and considered for performance comparison.
extending the BTC till level 3. Each plane will give the 8
elements of feature vector. For Red plane we get (UUUR,                          Figure 3 shows the average precision and recall values plotted
LUUR, ULUR, LLUR, UULR, LULR, ULLR, and LLLR). Also                              against number of retrieved images for Global BTC-CBIR at
feature vector of size 48 can be generated by extending this                     level 1 (BTC-6), level 2 (BTC-12), level 3 (BTC-24) and level 4
process of BTC to next level.                                                    (BTC-48), where the distinction in the performance of all these
                                                                                 techniques is not very clear.
6. CBIR using Local and Intermediate BTC
In the CBIR using local BTC, the feature vector is found by
using the local thresholding based BTC. Here the image is
divided into 2x2 pixel blocks, to find the local average of block.
This local average is treated as threshold value for generating the
bitmap of respective block, by comparing the local threshold
value with each pixel value of that 2x2 block. After completing
the bitmap for each plane of image the BTC can be applied to
generate the local feature for BTC-Level 1. This can be
continued for extracting the local multilevel BTC features using
BTC-Level 2, BTC-Level 3 and BTC-Level 4.

In CBIR using intermediate BTC, the either of intermediate-4 or                  Figure 3: Performance comparison of discussed Global thresholding based
intermediate-9 thresholding methods can be used to prepare                              BTC-CBIR methods using Precision-Recall crossover points
intial image bitmap, which further can be considered for
application of BTC to generate the intermediate BTC features of
various levels of BTC.

7. Implementation
The implementation of these CBIR techniques is done using
MATLAB 7.0. The CBIR techniques are tested on the
augmented Wang [17] image database of 1000 variable size
images spread across 11 categories of human beings, animals,
natural scenery and man-made things. To compare various
techniques in RGB color space, performance is evaluated based
on precision and recall. The efficiency of CBIR technique is
evaluated based on accuracy, stability and speed. To assess the
                                                                                   Figure 4:Crossover points of Precision-Recall plotted against number of
retrieval effectiveness, we have used the crossover point of                                  retrieved images for Global BTC-CBIR methods




                                                                          252                                   http://sites.google.com/site/ijcsis/
                                                                                                                ISSN 1947-5500
                                                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                             Vol. 09, No.06, June 2011



The height of crossover point of precision and recall curves
plays very important role in performance comparison of CBIR
methods. Ideally this crossover point height should be one.
Higher the value of this crossover point better the performance
is. Figure 4 shows the zoomed version of graphs in figure 3 for
crossover points of precision and recall curves. From figure 4 it
is evidently observed that BTC-level 3and level 4 outperform the
lower levels of BTC for CBIR using global thresholding.

Figure 5 and Figure 6 shows the bar graphs indicating the
heights of crossover points of precision-recall curves of
respective proposed BTC-CBIR methods.




                                                                                Figure 6: Performance comparison of discussed BTC-CBIR methods using
                                                                                        assorted thresholding methods for respective level of BTC


                                                                                9. Conclusion
                                                                                The BTC based CBIR methods have been actively researched
                                                                                and proven to be better. Even recently the multilevel BTC-CBIR
                                                                                techniques have been proposed for better image retrieval. But in
                                                                                all these BTC-CBIR methods the global thresholding was
                                                                                considered. The paper presented the appraise in performance of
                                                                                BTC-CBIR methods using intermediate thresholding methods.
                                                                                Here in all four assorted thresholding methods (alias local,
                                                                                intermediate-4, intermediate-9 and global) are used for BTC-
                                                                                CBIR in consideration with four levels of BTC from level 1 to
  Figure 5: Performance comparison of different levels of BTC-CBIR for          level 4, resulting into 16 variations of BTC-CBIR. The
                     respective thresholding methods
                                                                                experimental results have shown that the intermediate
                                                                                thresholding helps in performance improvement of BTC-CBIR
Figure 5 is serving the purpose of performance comparison of
                                                                                as compared to conventional global thresholding used in earlier
different levels of BTC for respective thesholding techniques
                                                                                versions. The BTC level 3 along with intermediate-9
proposed. From figure 5 it can be observed that for each
                                                                                thresholding method has given the best performance for image
thresholding method (except the intermediate-9 thresholding) the
                                                                                retrieval.
best performance (Highest precision-recall crossover point
value) is given by Level 4 (BTC-48). The increase in
performance from BTC-6 (Level 1) to BTC-24 (Level 3) is                         References
gradual in nature while the performances of BTC-24 (Level 3)                     [1]. Dr.H.B.Kekre, Sudeep D. Thepade, Shrikant P. Sanas
                                                                                      “Improved CBIR using Multileveled Block Truncation
and BTC-48 (Level 4) have very slight difference. This proves
                                                                                      Coding”, International Journal of Computer Science and
the worth of multilevel BTC over the conventional BTC (BTC
                                                                                      Engineering (IJCSE), Volume 2, Number 7, October 2010
level 1) with BTC level 4 being the best in CBIR.                                     Edition, pp2471-2476. http://www.enggjournals.com/ijcse.
                                                                                 [2]. Dr.H.B.Kekre, Sudeep D. Thepade, Shrikant P. Sanas
Figure 6 gives the performance comparison of proposed                                 “Improving Performance of Multileveled BTC Based CBIR
thresholding methods used in BTC for respective levels of BTC-                        Using Sundry Color Spaces”, International Journal of Image
CBIR. The best performance (Highest precision-recall crossover                        Processing (IJIP), Volume 4, Issue 6, Computer Science
point value) is given by Intermediate-9 thresholding of BTC-                          Journals, CSC Press, www.cscjournals.org
CBIR for all levels of BTC.                                                      [3]. Guoping Qiu, “Color Image Indexing Using BTC,” IEEE
                                                                                      Transactions on Image Processing, VOL.12, NO.1, pp.93-
So the overall conclusion that „to improve the performance of                         101, January 2003.
BTC-CBIR methods the combination of BTC-Level 3 (BTC-24)                         [4]. Dr.H.B.Kekre, Sudeep D. Thepade, “Boosting Block
with intermediate-9 thresholding has to be used‟ can be drawn                         Truncation Coding using Kekre‟s LUV Color Space for
based on the experimentation.                                                         Image Retrieval”, WASET International Journal of




                                                                         253                             http://sites.google.com/site/ijcsis/
                                                                                                         ISSN 1947-5500
                                                                       (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                          Vol. 09, No.06, June 2011

        Electrical, Computer and System Engineering (IJECSE),                         Thesis , Master of science in Informatics, Norwegian
        Volume 2, Number 3, pp. 172-180, Summer 2008. Available                       university of science and Technology, Department of
        online at http://www.waset.org/ijecse/v2/v2-3-23.pdf                          computer and Information science, June 2006.
[5].    Dr.H.B.Kekre, Sudeep D. Thepade, “Image Retrieval using               [16].   Rafael C.Gonzalez, Richard E. Woods,” Digital Image
        Augmented Block Truncation Coding Techniques”, ACM                            Processing,” University of Tennessee, Second Edition, ISBN
        International Conference on Advances in Computing,                            81-7808-629-8, Pearson Education Pvt. Ltd.,2002.
        Communication and Control (ICAC3-2009), pp. 384-390,                  [17].   http://wang.ist.psu.edu/docs/related/Image.orig        (Last
        23-24 Jan 2009, Fr. Conceicao Rodrigous College of Engg.,                     referred on 23 Sept 2008)
        Mumbai. Is uploaded on online ACM portal.                             [18].   Dr.H.B.Kekre, Sudeep D. Thepade, “Color Based Image
[6].    Dr.H.B.Kekre, Sudeep D. Thepade, “Scaling Invariant                           Retrieval using Amendment Block Truncation Coding with
        Fusion of Image Pieces in Panorama Making and Novel                           YCbCr Color Space”, International Journal on Imaging (IJI),
        Image Blending Technique”, International Journal on                           Volume 2, Number A09, Autumn 2009, pp. 2-14. Available
        Imaging (IJI), www.ceser.res.in/iji.html, Volume 1, No.                       online at www.ceser.res.in/iji.html
        A08, pp. 31-46, Autumn 2008. Dr. H.B. Kekre, Sudeep                   [19].   Dr.H.B.Kekre, Tanuja Sarode, Sudeep D. Thepade, “Color-
        Thepade & Shrikant Sanas International Journal of Image                       Texture Feature based Image Retrieval using DCT applied
        Processing (IJIP), Volume (4): Issue (6) 629                                  on Kekre‟s Median Codebook”, International Journal on
[7].    B.G.Prasad, K.K. Biswas, and S. K.Gupta, “Region–based                        Imaging (IJI), Volume 2, Number A09, Autumn 2009,pp.
        image retrieval using integrated color, shape, and location                   55-65. Available online at www.ceser.res.in/iji.html
        index”, computer vision and image understanding, Oct 2003.            [20].   M. Flickner, H. Sawhney, W. Niblack, J. Ashley, Q. Huang,
[8].    Minh N. Do, Member, IEEE, and Martin Vetterli, Fellow,                        B. Dom, M. Gorkani, Hafner, D. Lee, D. Petkovic, D. Steele,
        IEEE,” Wavelet-Based Texture Retrieval Using Generalized                      and P. Yanker. Query by image and video content: the QBIC
        Gaussian Density and Kullback-Leibler Distance,” IEEE                         system. IEEE Computer, 28(9):23–32, 1995.
        Transactions nn Image Processing, Vol.11, No.2, Feb 2002.             [21].   Dr.H.B.Kekre, Sudeep D. Thepade, A. Athawale, Adib
[9].    Dr.H.B.Kekre, Sudeep D. Thepade, “Rendering Futuristic                        Parkar, “Using Assorted Color Spaces and Pixel Window
        Image Retrieval System”, National Conference on                               Sizes for Colorization of Grayscale Images”, ACM-
        Enhancements in Computer, Communication and                                   International Conference and Workshop on Emerging
        Information Technology, EC2IT-2009, 20-21 Mar 2009,                           Trends in Technology (ICWET 2010), Thakur College of
        K.J.Somaiya College of Engineering, Vidyavihar, Mumbai-                       Engg. And Tech., Mumbai, 26-27 Feb 2010, uploaded on
        77.                                                                           online ACM Portal.
[10].   Dr.H.B.Kekre, Sudeep D. Thepade, “Using YUV Color                     [22].   Dr.H.B.Kekre, Sudeep D. Thepade, Shobhit W., Miti K.,
        Space to Hoist the Performance of Block Truncation Coding                     Styajit S., Priyadarshini M. “Image Retrieval with Shape
        for Image Retrieval”, IEEE International Advanced                             Features Extracted using Gradient Operators and Slope
        Computing Conference 2009 (IACC‟09), Thapar University,                       Magnitude Technique with BTC”, International Journal of
        Patiala, INDIA, 6-7 March 2009.                                               Computer Applications (IJCA), Volume 6, Number 8, pp.28-
[11].   Dr.H.B.Kekre, Sudeep D. Thepade, Archana Athawale,                            33,                      September                    2010.
        Anant Shah, Prathmesh Verlekar, Suraj Shirke,“Energy                          http://www.ijcaonline.org/volume6/number8/pxc3871430.pd
        Compaction and Image Splitting for Image Retrieval using                      f
        Kekre Transform over Row and Column Feature Vectors”,
        International Journal of Computer Science and Network
                                                                             Author Biographies
        Security (IJCSNS),Volume:10, Number 1, January 2010,
        (ISSN: 1738-7906) Available at www.IJCSNS.org.
                                                                                          Dr. H. B. Kekre has received B.E. (Hons.) in
[12].   Dr.H.B.Kekre, Sudeep D. Thepade, Archana Athawale,
                                                                                          Telecomm. Engineering. from Jabalpur University in
        Anant Shah, Prathmesh Verlekar, Suraj Shirke, “Walsh
                                                                                          1958, M.Tech (Industrial Electronics) from IIT
        Transform over Row Mean and Column Mean using Image
                                                                                          Bombay in 1960, M.S.Engg. (Electrical Engg.) from
        Fragmentation and Energy Compaction for Image
                                                                                          University of Ottawa in 1965 and Ph.D. (System
        Retrieval”, International Journal on Computer Science and
                                                                                          Identification) from IIT Bombay in 1970 He has
        Engineering (IJCSE),Volume 2S, Issue1, January 2010,
                                                                                          worked as Faculty of Electrical Engg. and then HOD
        www.enggjournals.com/ijcse.
                                                                                          Computer Science and Engg. at IIT Bombay. For 13
[13].   Dr.H.B.Kekre, Sudeep D. Thepade, “Image Retrieval using
                                                                                          years he was working as a professor and head in the
        Color-Texture Features Extracted from Walshlet Pyramid”,
                                                                                          Department of Computer Engg. at Thadomal Shahani
        ICGST International Journal on Graphics, Vision and Image
                                                                                          Engineering. College, Mumbai. Now he is Senior
        Processing (GVIP), Volume 10, Issue I, Feb.2010, pp.9-18,
                                                                                          Professor at MPSTME, SVKM‟s NMIMS (Deemed to
        Available                                            online
                                                                                          be University). He has guided 17 Ph.Ds, more than 100
        www.icgst.com/gvip/Volume10/Issue1/P1150938876.html
                                                                                          M.E./M.Tech and several B.E./B.Tech projects. His
[14].   Khalid Sayood ,” Introduction to Data Compression ,”
                                                                                          areas of interest are Digital Signal processing, Image
        University of Nebraska-Lincoln , Second Edition , ISBN:1-
                                                                                          Processing and Computer Networking. He has more
        55860-558-4, by Academic Press,2000.
                                                                                          than 320 papers in National / International Conferences
[15].    Stian Edvardsen,”Classification of Images using color,
                                                                                          and Journals to his credit. He was Senior Member of
        CBIR Distance Measures and Genetic Programming, “Ph.D.




                                                                      254                              http://sites.google.com/site/ijcsis/
                                                                                                       ISSN 1947-5500
                                                           (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                              Vol. 09, No.06, June 2011

IEEE. Presently He is Fellow of IETE and Life
Member of ISTE Recently ten students working under
his guidance have received best awards and two have
been conferred Ph.D. degree of SVKM‟s
NMIMS (Deemed to be 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 submitted Ph.D. Thesis to SVKM‟s NMIMS
(Deemed to be University), Mumbai. He has about 08
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 Management and
Engineering, SVKM‟s NMIMS (Deemed to be
University), Vile Parle(w), Mumbai, INDIA. He is
member of International Association of Engineers
(IAENG) and International Association of Computer
Science and Information Technology (IACSIT),
Singapore. He is reviewer for many international
journals. He has worked as member of International
Advisory Committee for many International
Conferences. His areas of interest are Image Processing
Applications and Biometrics. He has about 110 papers
in National/International Conferences/Journals to his
credit with a Best Paper Award at Int. Conference
SSPCCIN-2008, Second Best Paper Award at
ThinkQuest-2009 National Level faculty paper
presentation competition, Best Paper Award at
Springer Int. Conf. ICCCT-2010 and second best
research project award at Manshodhan-2011.

Shrikant P. Sanas has received B.E. (Computer)
degree from Mumbai University with First Class in
2008.Currently pursuing M-Tech from Mukesh Patel
School of Tech. Mgmt. and Engineering. SVKM‟s
NMIMS (Deemed to be University), Vile Parle(w),
Mumbai. Currently working as Lecturer in Ramrao
Adik Institute of Technology. Nerul, Navi Mumbai. He
has 02 papers in International Journals.




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