Steganalysis of Reversible Vertical Horizontal Data Hiding Technique

Document Sample
Steganalysis of Reversible Vertical Horizontal Data Hiding Technique Powered By Docstoc
					                                                               (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                              Vol. 8, No. 6, September 2010

   Steganalysis of Reversible Vertical Horizontal Data
                    Hiding Technique

       Thom Ho Thi Huong,                                    Canh Ho Van                                         Tien Trinh Nhat
Faculty of Information Technology,                 Dept. of Professional Technique,                        College of Technology,
   Haiphong Private University,                      Ministry of Public Security,                        Vietnam National University,
        Haiphong, Vietnam                                   Hanoi, Vietnam                                      Hanoi, Vietnam
       thomhth@hpu.edu.vn                              hovancanh@gmail.com                                    tientn@vnu.edu.vn




Abstract—This paper proposes a steganalysis scheme for                    to attack a specific steganography technique such as a
detecting the reversible vertical horizontal (RVH) data hiding [1].       steganalytic method was presented in [2] for detecting stego-
The RVH scheme was introduced in the IJCSIS International                 images using the method proposed in [3].
Journal Vol. 7, No. 3, March 2010. In the RVH data hiding, the
message bits are embedded into cover-image by two embedding                    In this paper, we proposed a steganalytic scheme to detect
phases: the horizontal embedding procedure HEm and the                    the RVH watermarking scheme introduced in brief in the
vertical embedding procedure VEm. The pixel pairs belonging to            abstract. Our experimental results show the feasibility of the
the horizontally embeddable and vertically embeddable pixel pair          proposed method. It is useful in detecting malicious activities
domain are transformed to mark message bits. Through analysis,            on stego-images and also suggests a design consideration for
we detect out that, the two histograms of LSB scanning                    future development of steganographic techniques. The rest of
horizontally and vertically vary from a stego-image to the cover          this paper is organized as follows. In the next section, we
image. Based on this observation, we design a specific steganlytic        present again the RVH scheme in brief. Section III describes
method for attacking the RVH steganography. Experimental                  the proposed steganalytic method. Experimental results are
results show the detection accuracies of the steganography with           given in section IV, and conclusions are made finally in
various embedding rates are acceptable. The proposed technique            Section V.
can be applied in detecting the misuse of steganographic
technology in malicious activities.                                               II.   REVIEW OF THE RVH DATA HIDING SCHEME
Keywords-Steganography, steganalysis, watermarking, cover image,              In the steganographic method proposed in [1] used the
stego image, payload, reversible data hiding.                             multiple embedding strategies to improve the image quality
                                                                          and the embedding capacity. Basically, this method embeds
                       I.    INTRODUCTION                                 each message bit b of the secret bit stream into each grayscale
    Steganography is [1, 3, 4, 5] is the art and science of               cover pixel pair of a grayscale cover image in raster scan order.
concealed communication. The basic concept is to hide the                 This scheme includes two main stages, namely, the horizontal
very existence of the secret message. Digital object such as a            embedding procedure HEm and the vertical embedding
text, image, video, or audio segment can be used as the cover             procedure VEm. For the HEm procedure, the input image is
data. To obtain acceptable hiding payload and keep fidelity of            horizontally scanned in raster scan order (i.e., from left to right
the stego-image, the LSB replacement techniques [2,4 or other             and top to bottom) to gather two neighboring pixels x and y
references] are popular and widely studied in the literature.             into a cover pixel pair (x, y). If y is an odd value, then the cover
These methods usually hide more data in image areas with                  pixel pair (x, y) is defined as a horizontally embeddable pixel
higher spatial variations. Reversible steganography [1,3-5] is            pair. Otherwise, the cover pixel pair (x, y) is defined as a
one of the interesting branches of steganographic technology in           horizontally non-embeddable pixel pair. For the VEm
which the original cover image can be reconstructed without               procedure, the input image is vertically scanned in raster scan
any loss.                                                                 order to group two neighboring pixels u and v into a pixel pair
                                                                          (u, v). If v is an even value, then the pixel pair (u, v) is defined
     Steganalysis is the counterpart of steganography, the goal           as a vertically embeddable pixel pair. Otherwise, the pixel pair
of the steganalysis is to detect the hidden message,                      (u, v) is defined as a vertically non-embeddable pixel.
equivalently, to discriminate the stego object from the non-
stego-object. The steganalysis techniques proposed in the                      The secret bit sequence S is divided into two subsequence
literature can be classified into two categories: the universal           S1 and S2. The bit stream B1 is created by concatenating the
steganalysis which is designed to detect the hidden message               secret subsequence S1 and the auxiliary data bit stream A1
embedded with various data embedding algorithms such as a                 (i.e., B1=S1||A1). Similarly, the bit stream B2= S2||A2. The
technique proposed in [6] is used to attack the LSB                       generation of A1 and A2 will be described latter. The overview
steganography, and the specific steganalysis which is designed            of the RVH embedding process is shown in Fig. 1.



                                                                      7                                http://sites.google.com/site/ijcsis/
                                                                                                       ISSN 1947-5500
                                                                    (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                   Vol. 8, No. 6, September 2010
                                                                             equal to S1 (i.e., B1=S1). During the execution of the HEm
                                                                             procedure, for the first LC1 pixels in O, when each pixel has
                                                                             been processed for embedding, its LSB is taken as an auxiliary
                                                                             data bit of A1 and appended to the end of B1. That is, B1 is
                                                                             gradually grown until the LC1 auxiliary data bits in A1 are
                                                                             concatenated into B1. Finally, the to_be_embedded bit stream
                                                                             is B1=S1||A1, which is completely embedded into O.
                                                                                 Similar to the generation of A1, the auxiliary data stream
                                                                             A2 is actually the LSBs of the first LC2 (LC2 is the length of
                                                                             the compressed location map CM2 ended with the unique end
                                                                             of map indicator EOM2) pixels in the image V and generated
                                                                             as follows. B2 initially equals the secret bit sequence S2.
         Fig. 1. Embedding phase of RVH steganographic system [1]            During the execution of the procedure VEm, for the first LC2
                                                                             pixels in the image U, when each pixel has been processed for
    Firstly, the bit sequence B1 is horizontally embedded into               embedding, its LSB is taken as an auxiliary data bit of A2 and
O by using the HEm procedure to obtain the output image T                    append to the end of B2 until the LC2 auxiliary data bits in A2
sized H x W pixels. Secondly, the compressed location map                    are concatenated into B2. Finally, the information bit sequence
CM1 whose length is LC1 (will be described later), is                        is B2=S2||A2, which is fully marked into the image U.
embedded in to T by using the least significant bit (LSB)
replacement technique to obtain the output image U with size                     For the purposes of extracting B1 and recovering O, a
of H x W pixels. Thirdly, the bit sequence B2 is vertically                  location map HL sized H x (W/2) is needed to record the
embedded into U by using the VEm procedure to get the output                 positions of the horizontally embeddable pixel pair (x, y) in O.
image V with size of H x W pixels. Fourthly, the compressed                  The location map HL is a one-bit bitmap. All the entries of HL
location map CM2 whose length is LC2 is embedded into V by                   are initialized to 0. If cover pixel pair (x, y) is the horizontally
using the LSB replacement technique to get the final stego                   embeddable pixel pair, then the corresponding entry of HL is
image with size of H x W pixels.                                             set to be 1. Next, the location map HL is losslessly compressed
                                                                             by using the JBIG2 codec (Howard et al, 1998 [8]) or an
    Each bit b in stream B1 is horizontally embedded into each               arithmetic coding toolkit (Carpenter, 2002 [7]) to obtain the
horizontally embeddable pixel pair (x, y) at a time by using the             compressed location map CM1 whose length is LC1. The
horizontal embedding rule HR defined below until the whole                   compressed location map CM1 is embedded into the image T
bit stream B1 is completely marked into O to obtain the output               by using the LSB replacement technique as mentioned above.
image T.                                                                     Similarly, for the purposes of extracting B2 and recovering the
   Each bit b in B2 is vertically embedded into each vertically              image U, we also need a location map VL sized (H/2) x W to
embeddable pixel pair (u, v) at a time by using the vertical                 mark the position of the vertically embeddable pixel pairs (u, v)
embedding rule VR defines below until the entire bit sequence                in U. Then, VL is also losslessly compressed by using the
B2 is concealed into U to get the output image V.                            JBIG2 codec or an arithmetic coding toolkit to obtain the
                                                                             compressed location map CM2 whose length is LC2. Next, the
   The horizontal embedding rule HR: For each pair (x, y), we                map CM2 is concealed into the image V by using the LSB
apply the following embedding rules:                                         steganography as mentioned above.
       HR1: If the to_be_embedded bit b=1, then the stego                      The final output of the embedding phase is the final stego
        pixel pair is unchanged by (x0, y0) = (x, y).                        image X with size of H x W pixels.
       HR2: If the to_be_embedded bit b=0, then the stego                    III.   THE PROPOSED STEGANALYTIC SCHEME FOR THE RVH
        pixel pair is changed by (x0, y0) = (x, y-1).
                                                                                                 STEGANOGRAPHY
   The vertical embedding rule VR: For each pair (u,v), we                       After embedding a large message sequence M (its ratio is
apply the following embedding rules:                                         about 90% of maximum embeddable capacity of image) into
       VR1: If the to_be_embedded bit b=0, then the stego                   the original image Baboon sized 512x512 pixels (show Fig. 2)
        pixel pair is unchanged by (u0, y0) = (u,v).                         using the RVH scheme to obtain the stego-image Baboon, we
                                                                             calculate histogram of the two images (cover Baboon image
       VR2: If the to_be_embedded bit b=1, then the stego                   and stego Baboon image), resulted in Fig. 3. It’s very hard to
        pixel pair is changed by (u0, y0) = (u, v+1).                        detect any difference between the two images.
    It is noted that the rule HR and VR don’t cause the                           However, when we separably calculate two histograms on
underflow and overflow problem. That is the changed pixel                    all pixel odd columns and all pixel even rows of the cover
pairs are assured to fall in the allowable range [0,255].                    Baboon image, shown in Fig. 4. Similarly, calculate two
                                                                             histograms on all pixel odd columns and all pixel even rows of
    The auxiliary data bit sequence A1 is actually the LSBs of
                                                                             the stego Baboon image, resulted in Fig. 5. It’s easy to
the first LC1 (LC1 is the length of the compressed location
                                                                             difference between pair histogram in Fig.4 (a) and Fig. 5 (a), in
map CM1 ended with the unique end of map indicator EOM1)
                                                                             Fig. 4 (b) and Fig. 5 (b). The informality appears in Fig 5 (a)
pixels in the image T and generated as follows. Initially, B1 is




                                                                         8                                http://sites.google.com/site/ijcsis/
                                                                                                          ISSN 1947-5500
                                                                           (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                          Vol. 8, No. 6, September 2010
and (b) due to embedding process of RVH scheme following
description in detail below.                                                         2000

                                                                                     1800

                                                                                     1600

                                                                                     1400

                                                                                     1200

                                                                                     1000

                                                                                      800

                                                                                      600

                                                                                      400

                                                                                      200

                                                                                        0
                                                                                            0      50         100        150       200        250        300      (a)
                                                                                     2500




                                                                                     2000
                  Fig. 2. The Baboon image sized 512x512 pixels
   4000
                                                                                     1500
   3500


   3000
                                                                                     1000
   2500


   2000
                                                                                      500
   1500


   1000
                                                                                        0
                                                                                            0      50         100        150        200        250          300
    500                                                                                                                                                           (b)
      0                                                                              Fig. 5. Histogram of the stego Baboon images: (a) histogram on all pixel odd
          0       50       100       150      200       250       300
                                                                         (a)
                                                                                                   columns, (b) histogram on all pixel even columns
    4000


    3500


    3000
                                                                                       According to the horizontal embedding procedure HEm,
    2500
                                                                                    from an input image O, the pixels of the image O are
    2000
                                                                                    horizontally grouped into pixel pairs (x, y), these pairs are
    1500
                                                                                    partitioned into two sets, one set is E1 and other set is , the
    1000
                                                                                    set E1 contains pixels pair which are horizontally embeddable
     500
                                                                                    pixel pairs, while the set   consists of those pixel pairs which
          0
                                                                                    are horizontally non-embeddable pixel pairs.
              0    50       100      150       200       250       300   (b)
                                                                                       Now, we examine the migration of LSB histogram of the
  Fig. 3. Histogram of the tested two images: (a) the cover Baboon image,
                         (b) the stego Baboon image                                 image O and the image T obtained after embedding secret bit
   2000
                                                                                    B1. Without loss of generality, let (x, y) and (           be the
   1800

   1600
                                                                                    corresponding pixel pairs in the image O and the stego-image
   1400
                                                                                    T, respectively. In the horizontal embedding procedure HEm,
   1200                                                                             pixel pairs (x, y) E1, i.e. the LSB of pixel y be bit 1, are
   1000
                                                                                    selected to embed message bits. Here, We don’t examine
   800

   600
                                                                                    change of LSB histogram of pixels x on pixel even-columns
   400
                                                                                    because they are still remained value after embedding message
   200                                                                              bits. In the image T, the LSB of is changed to either 0 or 1,
     0
          0       50       100       150       200       250       300   (a)        and each of them appears in the same probability. It is
  2000
                                                                                    obviously that the probability of bit 0 and bit 1 are 0.5 and 0.5,
  1800                                                                              respectively. For pixel pairs (x, y)  , i.e. the LSB of pixel y
  1600                                                                              be bit 0, after embedding secret bits, is unchanged. So the
  1400
                                                                                    probability of bit 0 and 1 are 1 and 0, respectively.
  1200

  1000                                                                                 Next, the compressed location map CM1 (CM1 is a binary
   800
                                                                                    stream, whose length is LC1) are marked into the image T by
   600
                                                                                    the LSB replacement technique to obtained image U. That
   400

   200
                                                                                    changes a part of probability of LSB of bit 1 and bit 0 on all
     0
                                                                                    pixel even-columns in the image T. Assume that the bits are
         0        50       100       150       200       250       300   (b)
                                                                                    randomly distributed, so the probability of bit 0 and bit 1 are
Fig. 4. Histogram of the cover Baboon images: (a) histogram on all pixel odd        Pmap1(0) = Pmap1(1).
              columns, (b) histogram on all pixel even columns




                                                                                9                                    http://sites.google.com/site/ijcsis/
                                                                                                                     ISSN 1947-5500
                                                               (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                              Vol. 8, No. 6, September 2010
   Base on above discussions, the probabilities of bit 0 and bit           used to embed message, i.e. the embedding ration of P R-
1 of all pixels on even-column in the image U can be                       H=0.45       0.9=0.405. From (1), we have PLSB-H(0)
calculated. Assume the probability of pixel pairs belonging to             =0.405 (0.5 0.45+0.55) + 0.595 0.5 = 0.611375 and PLSB-
E1 and the probability of pixel pairs belonging to        be               H(1)=0.405     (0.5    0.45)+0.595 0.5 = 0.388625. Next,
and     , respectively. After marking the location map CM1,                calculated the probability of bit 0 and the probability of bit 1 of
PE1 and       are changed to P’E1 and         . Let PR-H is the            the output image X. We know that the probability of E2 equals
embedding ratio defined by dividing the number pairs actually              to the probability of the LSB of bit 0 of all pixels on even –
used to hide data by the total number of pairs the image O. The            rows, i. e. PE2 = PLSB(0)/2 + PLSB_H(0)/2 = (0.5+0.611375)/2 =
probability of bit b={0,1} of LSB of the image U can be                    0.5556875 and        =0.4443125. After covering a part of the
calculated using the following equation                                    LSB of image V by the location map CM2 with a probability
                                                                           0.05 (assumed), the two probability PE2 and             change to
                                                                (1)        P’E2=0.5056875 and        = 0.4943125, respectively.

   For the vertical embedding procedure VEm, vertically scan                  The embedding ratio of 90 % of the embeddable pairs are
the output image U in raster scan order to group pixel pairs (u,           used to embed message, i.e. the embedding ration of P R-
v), we classify the pairs into two sets, one is E2 and other is            V=0.5056875 0.9=0.45511875. So probability of LSB of bit 0

    , the set E2 contains all pixel pairs which are vertically             and bit 1 of the output image X from (2) we obtain PLSB-
embeddable pixel pairs, the set         consisting pixel pairs are         V(0)=0.45511875     (0.5 0.5056875) + 0.54488125       0.5
vertically non-embeddable pixel pairs. Let (u, v) and (         be         0.3875, PLSB_V(1)0.61248.
pixel pairs of the image U (before using the procedure Vem)                   Now, we check again the probability of LSB of bit 1 and the
and the stego-image V (after embedding secret message using                probability of LSB of bit 0 of all pixels on pixel even-columns,
the procedure Vem). In the procedure VEm, only pixel pairs (u,             PLSB_even_column(0)=PLSB_H(0)/2 + PLSB_V(0)/2 =(0.611375+
v) E2, i.e. the LSB of v is bit 0, are embedded message bits.              0.3875)/2=0.4994375, PLSB_even_column(1) = (PLSB_H(1) +
After embedding message bits, the LSB of the pixels                        PLSB_V(0))/2=(0.388625+0.61248)/2=0.5005525. We found out
(obtained from E2) is either 0 or 1. So, the probabilities of bit 0        that the probability of bit 0 PLSV_even_column(0) and probability of
and bit 1 of (     are equals to 0.5. For pixel pairs (u, v)  ,           bit 1 PLSB_even_column(1) are the same, that is after completing the
i.e. the LSB of pixel v be bit 1, after embedding secret bits, is          vertical procedure VEm, it make the value of these
unchanged. So the probability of bit 0 and 1 are 0 and 1,                  probabilities be balanced. However, the probability of LSB of
respectively.                                                              bit 0 and bit 1 of all pixels on pixel odd-columns don’t equal
                                                                           based           on         the         following          calculating:
  Next, the compressed location map CM2 are marked into the
                                                                           PLSB_odd_column(0)=(PLSB_org_odd_column(0)/2+PLSB_V(0)/2)=(0.5/2 +
image V by the LSB replacement technique to obtained image
                                                                           0.3875/2)=0.44375, PLSB_odd_column(1)=(PLSB_org_odd_column(1)/2 +
X. That changes a part of probability of LSB of bit 1 and bit 0
                                                                           PLSB_V(1)/2)=(0.5/2+0.61248/2)=0.55624.                        Where
on all pixel even-rows in the image V. Assume the bits of the
                                                                           PLSB_org_odd_column(0) and PLSB_org_odd_column(1) be the probabilities
map CM2 are randomly distributed, so the probability of bit 0
                                                                           of the LSB of bit 0 and bit 1 of all pixels on the pixel odd-
and bit 1 are Pmap2(0) = Pmap2(1).
                                                                           column of the image X. A half of them isn’t changed during
   From above discussions, the probabilities of bit 0 and bit 1            process of the RVH scheme, so PLSB_org_odd_column(0)/2 and
in the LSB of image X after using the vertical embedding                   PLSB_org_odd_column(1)/2 equal to 0.5/2 and 0.5/2, respectively. We
procedure VEm can be calculated. Assume the probability of                 can see obvious difference of the occurrences of bit 1 and bit 0
pixel pairs belonging to E2 and the probability of pixel pairs             in the LSB on all pixel odd_columns and all pixel
belonging to      be     and     , respectively. After marking             even_column of the stego-image of the RVH scheme with
the location map CM2, PE2 and         are changed to P’E2 and              respect to a standard natural image. Based on the problem, the
     . Let PR-V be the embedding ratio defined by dividing the             following rule is given to discriminate a stego-image of the
number pairs actually used to hide data by the total number of             RVH steganography from a nature image.
pairs the image V. The probability of bit b={0,1} of LSB of
the image V can be calculated using the following equation                    W(X) =                                                            (3)

                                                               (2)            From equation (3), an image is detected be stego-image
                                                                           marked by the RVH scheme if one of the measured values
    For a natural image, assume that the LSB is randomly                   |PLSB(0) – PLSB(1)| on all pixel odd columns (or pixel even
distributed, then the expected probability of bit 0 and the                columns) or pixel even rows (or pixel odd rows) is greater than
probability bit 1 of all pixel on pixel even-columns are the               threshold T (0 ≤ T ≤ 1). The threshold T is used to control the
same, i.e. PLSB(0) = PLSB(1)=0.5. So probability PE1=0.5,                  decision boundary of nature images and stego images, its value
     =0.5. After covering a part of the LSB of image T by the              depends on specific applications.
location map CM1 with a probability 0.05 (assumed), the two
probability PE1 and        change to P’E1=0.45 and     =0.55,                                 IV.   EXPERIMENTAL RESULT
respectively. Consider the Baboon stego-image from the                        To show the reliability of the proposed method, we take
Baboon cover image, the probability of the embeddable pairs                500 image from USC-SIPI Image Database [9] and content
(i.e. those pixel pairs belonging to the procedure HEm) of an              based image retrieval (CBIR) image database [10] and convert
input image T is P’E1, and 90 % of the embeddable pairs are                them into 8-bit grayscale images. The images are used to test



                                                                      10                                 http://sites.google.com/site/ijcsis/
                                                                                                         ISSN 1947-5500
                                                                                             (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                            Vol. 8, No. 6, September 2010
proposed classification. The hidden messages used in our test
are made by the pseudo random number generator. We embed
                                                                                                                                              0.4
different amount of message using the RVH scheme, and
measure the migration of LSB histogram in the stego images.                                                                          0.35
Five embedding ratios 0%, 25%, 50%, 75% and 100% are used
                                                                                                                                              0.3
in the test, and the obtained |PLSB(0)-PLSB(1)| values on all pixel
even-rows of the stego-images are depicted in Fig. 6 – 10,




                                                                                                       |Plsb(0) - Plsb(1)|
                                                                                                                                     0.25
respectively. We also measure the accuracy of the proposed
method in detecting the RVH scheme in different embedding                                                                                     0.2

ratio and likelihood threshold value T, shown in table 1.                                                                            0.15

                          0.25
                                                                                                                                              0.1


                                                                                                                                     0.05
                           0.2

                                                                                                                                               0
                                                                                                                                                    0         50     100   150       200   250    300    350     400   450     500
    |Plsb(0) - Plsb(1)|




                          0.15                                                                                                                                                       Number of images


                                                                                                      Fig. 9. The distribution of |PLSB(0) – PLSB(1)| value of the 500 stego images on
                           0.1                                                                                         all pixel even-rows with embedding ratio 75%

                                                                                                                                               0.5
                          0.05
                                                                                                                                              0.45

                                                                                                                                               0.4
                             0
                                 0   50   100   150   200   250    300   350   400   450   500
                                                                                                                                              0.35
                                                      Number of images
                                                                                                                        |Plsb(0) - Plsb(1)|
                                                                                                                                               0.3
Fig. 6. The distribution of |PLSB(0) – PLSB(1)| value of the 500 cover images on
                               all pixel even-rows                                                                                            0.25

                                                                                                                                               0.2
                          0.35

                                                                                                                                              0.15
                           0.3
                                                                                                                                               0.1

                          0.25                                                                                                                0.05
 |Plsb(0) - Plsb(1)|




                                                                                                                                                    0
                           0.2                                                                                                                          0      50    100       150   200   250    300    350     400   450   500
                                                                                                                                                                                     Number of images

                          0.15
                                                                                                      Fig. 10. The distribution of |PLSB(0) – PLSB(1)| value of the 500 stego images on
                                                                                                                      all pixel even-rows with embedding ratio 100%
                           0.1

                                                                                                      TABLE I.                                                  THE DETECTION ACCURACY OF THE PROPOSED METHOD WITH
                          0.05                                                                                                                              VARIOUS EMBEDDING RATIOS AND THRESHOLD VALUES

                            0                                                                                                                   Embedding
                                 0   50   100   150   200   250    300   350   400   450   500
                                                      Number of images
                                                                                                                                                 Ratio (%)                 0           25         50            75       100
Fig. 7. The distribution of |PLSB(0) – PLSB(1)| value of the 500 stego images on                                              Threshold T
                 all pixel even-rows with embedding ratio 25%                                                                                                Cover     70.6 %          2%        0.6%          0%        0%
                                                                                                                              0.01
                                                                                                                                                             Stego     29.4 %         98%       99.4 %         100%     100 %
                          0.35                                                                                                                               Cover     80.6 %        5.8 %       0.6 %          0%        0
                                                                                                                              0.02                           Stego     19.4 %        94.2 %     99.4 %         100%     100 %
                           0.3                                                                                                                               Cover     84.4 %        7.4 %       2.2 %           0        0
                                                                                                                              0.03                           Stego     15.6 %        92.6 %     97.8 %         100%     100%
                          0.25                                                                                                                               Cover     86.8 %         11 %       2.4 %           0        0
                                                                                                                              0.04                           Stego     13.2 %         89 %      97.6 %         100%     100%
  |Plsb(0)-Plsb(1)|




                           0.2
                                                                                                                                                             Cover      89 %         13.8 %      3.6 %         0.2 %      0
                                                                                                                              0.05                                                   86.2%      96.4 %         99.8%    100%
                          0.15                                                                                                                               Stego      11 %

                           0.1

                                                                                                          From Fig. 6 , we can see that most of the value of |PLSB(0) –
                          0.05
                                                                                                      PLSB(1)| approach zero for natural images, while the higher
                             0                                                                        value of |PLSB(0) – PLSB(1)| is obtained with embedding ratio
                                 0   50   100   150   200   250    300   350   400   450   500
                                                      Number of images                                25%, 50%, 75% and 100% shown from Fig. 7 to Fig. 10. From
                                                                                                      table 1, we see that when the likelihood threshold value T is set
Fig. 8. The distribution of |PLSB(0) – PLSB(1)| value of the 500 stego images on
                 all pixel even-rows with embedding ratio 50%




                                                                                                 11                                                                                  http://sites.google.com/site/ijcsis/
                                                                                                                                                                                     ISSN 1947-5500
                                                                         (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                        Vol. 8, No. 6, September 2010
0.035, we can obtain an acceptable result in detecting stego                          [7]  Carpenter, B., 2002. Compression via Arithmetic Coding
images used the RVH steganography.                                                         http://www.colloquial.com/ArithmeticCoding/
                                                                                      [8] P.G. Howard, F. Kossentini, B. Martins, S. Forchhammer, W. J.
                            V.     CONCLUSION                                              Rucklidge, 1998. The emerging JBIG2 standard. IEEE Transactions on
                                                                                           Circuits and Systems for Video Technology 8 (7), pp. 838-848.
   The paper presents a method to break the RVH                                       [9] USC-SIPI                            Image                      Database,
steganography based on the observation of the distribution of                              “http://sipi.usc.edu/services/database/Database.htm”
0 and 1 bits of the LSBs on pixel odd-columns (pixel even-                            [10] CBIR         Image      Database,      University    of   Washington,
columns) or pixel even-rows (pixel odd-rows) of the RVH                                    http://www.cs.washington.edu/research/imagedatabase/groundtruth/..
stego-images. The experimental results are shown that the
proposed method can detect stego - images reliably with                                                          AUTHORS PROFILE
embedding ratio being greater 25%. On the other hand, we
show a problem of security of the RVH scheme in the data                                                    Ho Thi Huong Thom received the B.S. degree of
embedding process.                                                                                          Information Technology department from Haiphong
                                                                                                            Private University and the M.S. degree in Information
                          ACKNOWLEDGMENT                                                                    Systems from College of Technology, Vietnam National
    Our special thanks to Haiphong Private University (HPU)                                                 University in Vietnam, in 2001 and 2005, respectively.
for their financial support to our research and College of                            She has started her career as Lecturer in Department of Information
Technology, Vietnam National University, Hanoi for their                              Technology in Haiphong Private University, Vietnam and served for 9 years.
support to good working environment. We would like to                                 Currently, she is pursuing Doctor of Information Systems from College of
extend our thanks to my guide, our friends and family
                                                                                      Technology, Vietnam National University, Hanoi, Vietnam. Her research
members without whose inspiration and support our efforts
                                                                                      interests includes Image processing, Information Security, Information Hiding.
would not have come to success.
                               REFERENCES                                                                 Ho Van Canh received the B.S. degree in Mathematics
                                                                                                          from Hanoi City University in Vietnam in 1973, the Dr.
[1]   P. Mohan Kumar, K. L. Shunmuganathan, A reversible high embedding                                   Sci. degree in Faculty of statistology from KOMENSKY
      capacity data hiding technique for hiding secret data in images,
      International Journal of Computer Science and Information Security,                                 University in Czechoslovakia in 1987. Currently, he has
      Vol.7, No. 3, March 2010, pp. 109-115.                                                              been working as a cryptologist in Dept. of Professional
[2]   Yeh-Shun Chen, Ran-Zan Wang, Yeuan-Kuen Lee, Shih-Yu Huang,                                         Technique, Ministry of Public Security, Vietnam. His
      Steganalysis of reversible contrast mapping water marking, Proceedings
      of the world congress on Engineering 2008 Vol I, WCE2008, July 2-4,             research interests include cryptography, information security, information
      2008, London, U.K., pp. 555-557.                                                hiding.
[3]   D. Coltuc and J. M. Chassery, “ Very fast watermarking by reversible
      contrast mapping,” IEEE Signal Processing Lett., vol. 14, no. 4, pp.
      255– 258, Apr. 2007.                                                                                Trinh Nhat Tien received the B.S degree from University
[4]   J. Tian, “ Reversible Data embedding using a difference expansion,”                                 of Prague in Czechoslovakia in 1974, and the Dr. degree
      IEEE Trans. Circuits Syst. Video technol., vol. 13, no. 8, pp. 890– 896,
      Aug. 2003.                                                                                          from University of Prague, Czechoslovakia and University
[5]   Z. Ni, Y. Q. Shi, N. Ansari, and W. Su, “ Reversible Data Hiding,”IEEE                              of Hanoi, Vietnam in 1984. He has started as Lecturer in
      Trans. Circuits Syst. Video technol., vol. 16, no. 3, pp. 354– 362, Mar.                            Department of Information Technology of College of
      2006.
[6]   J. Fridrich, M. Goljan, and R. Du, “ Reliable detection of LSB                  Technology, Vietnam National University, Hanoi, Vietnam since 1974. His
      steganography in color and grayscale images,”Proceedings of the ACM             research interests include algorithm, complexity of algorithm, information
      International Multimedia Conference and Exhibition, pp. 27– 30, 2001.
                                                                                      security.




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

				
DOCUMENT INFO
Description: IJCSIS is an open access publishing venue for research in general computer science and information security. Target Audience: IT academics, university IT faculties; industry IT departments; government departments; the mobile industry and computing industry. Coverage includes: security infrastructures, network security: Internet security, content protection, cryptography, steganography and formal methods in information security; computer science, computer applications, multimedia systems, software, information systems, intelligent systems, web services, data mining, wireless communication, networking and technologies, innovation technology and management. The average paper acceptance rate for IJCSIS issues is kept at 25-30% with an aim to provide selective research work of quality in the areas of computer science and engineering. Thanks for your contributions in September 2010 issue and we are grateful to the experienced team of reviewers for providing valuable comments.