CHAPTER 7 Steganography based on Random pixel selectionshodhganga.inflibnet.ac.in/bitstream/10603/97221/9/th-1816_ch7.pdf · image based steganography is selecting of technique that

126

CHAPTER 7

Steganography based on Random pixel

selection

7.1 Introduction

An ideal Steganography technique embeds data into images in a way that forms

modified images which are visually imperceptible. As the application domain of

embedding data in image source becomes broaden, data hiding techniques are

designed in terms of security, capacity and imperceptibility. The performance of a

steganographic system can be measured using some properties of which the most

important property is the statistical undetectability of the data, which exhibits

difficulty in determining the existence of a hidden message. Another associated

measure is the steganographic capacity, where the maximum information can safely

embedded in a cover-image without having statistically detectable objects (Hamid et

al., 2013).

LSB based technique enables high capacity of data embedding and also maintaining

the visual perception of the stego-image. This method is probably the easiest way of

hiding information in an image and yet it is surprisingly effective (Johnson and

Jajodia, 1998). As stated by Jain et al., (2012b) LSB steganography is described as

follows: if the LSB of the pixel value I(i, j) is equal to the message bit m to be

embedded, I(i, j) remain unchanged; if not, set the LSB of I(i, j) to m. The message

embedding procedure is described using equation 7.1 as stated below,

(7.1)

where LSB(I(i, j)) stands for the LSB of I(i, j) of the cover-image, m is the bit to be

embedded and IS (i, j) is the pixel‟s LSB value of the stego-image.


selection

127

Such approach employs lossless compression in its mechanism that preserves the

quality and also increases the quantity of the secret data. LSB methods typically

achieve high capacity but unfortunately LSB insertion is vulnerable to attacks (Pujari

and Mukhopadhyay, 2012). In LSB steganography framework, the technique for

embedding the message may be known to an attacker and so it is generally not

considered as good practice to rely on the secrecy of the algorithm itself. The simple

LSB method can be enhanced with often much more sophisticated approach.

A digital image is defined as an arrangement of numbers and such numbers usually

stand for different light intensities in different parts of the image. In a color scheme,

the number of bits represents the bit depth and this basically refers to the number of

bits assigned to each pixel (Morkel et al., 2005).

Figure 7.1: RGB color space showing primary colors and gray value

The RGB (Red-Green-Blue) color model has three basic primary (basic) colors: red,

green and blue. All other colors are obtained by combining them (Carvajal-Gamez et

al., 2009; Kaur et al., 2010). In RGB color space, the colors with „P‟ are the primary

colors and the dashed line indicates where to find the grays, going from (0, 0, 0) to

(255, 255, 255) as shown in the figure 7.1. Images consist of pixels with contributions

from primary colors (red, green and blue) adding to the total color composition of the

pixel (Johnson and Jajodia, 1998). Each pixel typically has three numbers associated

with it, one each for red, green, and blue intensities, and these value ranges from 0-

255 (Gonzalez and Woods, 2002; Hosmer, 2006). Depending on the depth of color

desired in the final image, each component is represented by a separate number of

bits. When represented as decimal contributions, a value of (255, 0, 0) would describe

a 100% red pixel. By mixing the contribution of each component different colors can


selection

128

be represented. Value mixtures such as (31, 187, 57) can result in a dark green while

(255, 255, 0) represents pure yellow.

(255, 255, 0) (254, 255, 0)

Figure 7.2: Red plane LSB Modification

When any specific color is viewed closely, single digit modifications to the

contribution level are imperceptible (i.e. a pixel with a value of (255, 255, 0) is

indistinguishable from (254, 255, 0)) as shown in figure 7.2 that illustrates the impact

of modifying one bit in the red contribution. One of the most important properties of

image based steganography is selecting of technique that can fulfill the basic

requirements in developing efficient steganographic process. If the LSB‟s of the red

plane is changed only, then such change does not have a noticeable impact on the

color of a pixel as seen in figure 7.2.

The LSB based data embedding can differ in the way of data hiding process and

different variations of LSB insertion exist. Information bits can be embedded in

image‟s LSB sequentially or randomly at fixed place or by random scattering

throughout the cover-image (Cheddad et al., 2009; Hamid et al., 2013). In the

sequential embedding scheme, the LSBs of the image are replaced by the message bit

sequentially (Tomar, 2012). Embedding depends on the length of the message and

thus the position of the pixel where the secret message is hidden is modified while the

rest remain untouched. But the disadvantage of sequential embedding is that the

message is encoded into the cover-image sequentially (ordered), so the clusters of bits

embedded can be found that results in abrupt changes in the bits statistics and this can

result in detection (Zhang et al., 2006).

For random case, there are no such clusters because the embedded bits are scattered

into the cover-image, so the detection process is not expected to be as effective as for

the sequential case. Randomizing the embedding process is one of the variations of

LSB approach where the secret bits can be placed into the pre-selected pixels of the


selection

129

cover-image. Thus the bits of secret message bits will not be beside in one part, but

instead randomly dispersed throughout the image. In such embedding scheme, the

message bits are spread throughout the whole image using a random sequence to

control the embedding sequence (Hamid et al., 2013; Tomar, 2012). By randomizing

the embedding approach, the method to estimate the statistical properties of the image

used can be effectively disabled. Such approach of hiding in a randomized manner is

quite appealing and proves effective. In this process, a pseudorandom number

generator (PRNG) is used to select random pixels of the cover-image for hiding the

secret data. A chosen key can be inserted into a PRNG which will determine a

sequence of random numbers (Manchanda et al., 2007). These numbers indicate the

pixels in the cover-image where the LSBs are to be changed. In such case without the

secret key, it becomes significantly difficult for the attacker to figure out the target

pixels (Weng et al., 2009; Cole and Krutz, 2003). This makes the system more secure

because the receiver of the message must know the key in order to determine the

locations of the bits of the hidden message. So, in case the stego-image is known,

embedding the message in a random manner would not affect the security as long as

the stego-key remains unknown to the attacker. The secret key is the password used to

seed a pseudo-random number generator to select pixel locations in an image cover-

image for embedding the secret message.

Wu and Tsai, (2003), proposed a steganographic method based on pseudorandom

mechanism on gray-scale images by using pixel value differencing method where the

number of bits to be embedded in a pixel pair is decided by the difference value. The

PSNR value ranges from 40-47 dB for embedding average of 50960 bits. According

to Liang et al., (2007), pseudo-random based data embedding avoids concentration of

pixel alterations and ensures data security. They proposed a scheme that encodes the

secret bits directly into boundary pixels by checking each pixel of the cover image in

a pseudo-random order for embedding eligibility. Luo et al., (2010), proposed an edge

adaptive scheme which selects the embedding regions according to the size of secret

message and the difference between two consecutive pixels in the cover image

according to a pseudorandom number generator (PRNG). They stated that such

approach can resist both visual and statistical attacks and shows better PSNR value.

Jain et al., (2012b), presents random LSB based steganography where the edge pixels

were selected to hide the secret data. They stated that the selection of pixels for


selection

130

embedding is crucial because modified pixels in areas of the image where there are

pixels that are most like their neighbors are much more noticeable to the naked eye

and to solve this problem edge pixel were randomly selected to embed the message.

Amirtharajan et al., (2013), proposed randomized LSB embedding based on some

routines, in which the method embeds data into the selected pixels based on the four

LSBs of the Key. If key K = 13[0 0 0 0 1 1 0 1] then embedding should be in the 1st,

3rd

and 4th

LSBs. If key K = 165 [1 0 1 0 0 1 0 1] then embedding should be done in

the pixels 1st and 3

rd LSBs. This sequence of embedding is repeated for a block of

eight pixels for complete embedding. But there exists problem with such approach in

that if the LSBs of the key value are all zeroes then no embedding can take place.

For all the methods discussed above embeds the secret data in all the three planes

(Red, Green and Blue) in random manner.

7.2 Proposed Data Hiding Technique

This chapter introduces an approach of least significant bit (LSB) based

steganography in digital images that can override some statistical and structural

measures of detection. In this method, there is a provision of increasing the robustness

by spreading message bits randomly. One of the main characteristics of this

steganographic technique is that the data are embedded only in the red plane of the

cover-image‟s pixel which is determined by a PRNG (Pseudo-Random Number

Generator) initiated by a stego-key. A key is used as seed for the pseudo-random

number generator which is needed in the embedding process (Venkatraman et al.,

2004). By using the key, the chance of getting attacked by the third party is reduced as

it spreads the secret data over the whole stego-image randomly. Such mechanism

preserves the visuality of the stego-image finely so that the manipulation is not

detectable by visual interpretation. In general the embedding process inserts a mark X

(secret message), in an object Y (cover-image). A key K, initiated by a random

number generator is used in the embedding process and the resulting marked object Y

is generated by mapping

X * Y * K = Y (7.2)


selection

131

The pseudorandom number generator is used to choose the pixels randomly and

embed the message into the selected pixels that makes the message bits more difficult

to find and reduces the realization of patterns in the image (Luo et al., 2011;

Manchanda et al., 2007). If the message is much smaller than the capacity of the

image, a problem may occur whereby the information is packed into one part of the

image for example the top half. This is solved by using a PRNG which will spread the

message all over the image (Dedic et al., 2009). A pseudo random number generator

calculates and selects the order of pixels to be chosen for data embedding based on

the key. This randomization function is reversible function, i.e. the function generates

a random sequence of a bit stream input, but it can also regenerate the original

sequence when the randomized sequence is taken as input (Pujari and Mukhopadhyay,

2012). Thus, data is hidden into the LSB of the red plane of the randomly selected

pixel of the cover-image using the mechanism of PRNG which further enhances the

process of randomization.

7.2.1 Data embedding procedure

In the embedding process the pixels of the cover-image are selected randomly and

their red plane‟s LSB is replaced with the bits of the secret message. As image

comprises of pixel contribution from color components and changing the LSBs does

not affect the perception of the resulting image. For 24-bit bitmap image if a pixel‟s

red plane value is 10111011and changing the LSB with the secret bit, then at the

worst case the pixel value changes to 10111010, which does not affect the statistics of

the image. Changing the least bit in a byte of an image, either increment or decrement

value from the pixel it represents. So, the secret messages are stored into LSBs of red

plane of the cover-image‟s pixels.

In performing the hiding operation, the secret message is first converted into byte

format and stored in a byte array. It is then embedded into the LSB position of each

pixel‟s red plane. It uses the first pixel (at spot 0) to hide the length of message

(number of character). If the original pixel bits is considered as: (r7 r6 r5 r4 r3 r2 r1

r0, g7 g6 g5 g4 g3 g2 g1 g0, b7 b6 b5 b4 b3 b2 b1 b0) and a character of the secret

message has bits: (c7 c6 c5 c4 c3 c2 c1 c0). The character bit is embedded into the

LSB of red plane only while keeping LSBs of the green and blue plane remain


selection

132

unaltered. So, after embedding the character bit the new pixel is formed as (r7 r6 r5 r4

r3 r2 r1 c0, g7 g6 g5 g4 g3 g2 g1 g0, b7 b6 b5 b4 b3 b2 b1 b0). In this way, the next

character bit is embedded into the LSB of the next randomly selected pixel‟s red plane

and the process continues until the entire message is embedded. Thus, it ensures that

the secret message is embedded into different LSBs of the selected pixel‟s red plane

only. So selected pixels used for embedding are scattered and thus increases the

security of message highly.

If a pixel (225,100,100) represented in binary form (11100001, 01100100, 01100100)

is a part of the cover-image into which message character “d” having binary value

1100100 (ASCII value 100) is embedded then placing the first bit of “d” in pixel‟s red

plane a new pixel is obtained as (224, 100,100) represented in binary (11100000,

01100100, 01100100). Here it is noticed that the pixel value of (225, 100,100) is

changed to (224,100,100). From experiments it is observed that such changes will not

have noticeable statistical and color difference in the image.

In this method the red plane is selected for data embedding. The blue and green planes

are left unmodified. So after performing the embedding operation the unmodified

green and blue planes are added to the modified red plane to form the final stego-

image.

7.2.2 Key based Random embedding

In the process of LSB embedding, pseudo random number generator (PRNG) is used

to select the hiding points in the red plane of the cover image pixel by using a random

interval method. One of the most basic primitive of pseudorandom number generator

is the random generator, also known as a keystream generator. Here a random

generator is used that randomly select pixel‟s positions to hide the bits of a secret

message into the LSB of the red plane within a cover image. When using an

algorithmic keystream generator is to have a seed as well as a key. The keystream in

order to appear “random” must have the property an opponent who gets their hands on

even a number of keystream bits should not be able to predict any more of them. A

pseudo-random sequence is generated from a pseudo-random number generator

(PRNG) initialized by a secret stego-key. The transmitting and receiving end share the

stego-key, the output of the key is a random sequence K1, K2,……….., Kn where n is


selection

133

the length of message bits. Thus key is used not only as password but also as

sequence of pseudorandom permutations which locates the random pixel positions in

a cover-image. The pseudo-random sequence can be started at any position in the

sequence by initializing the generator with a seed. The seed defines the starting

position of the sequence and based on the seed the key produces a long random

sequence which is then used by the sender to generate the sequence of pixel indices yi

to get random location where,

y1 = K1

yi = yi-1 + Ki, i >2. (7.3)

The inputs to the embedding process are the bits of pixel indices Y and key K. Each

round has inputs Yi-1 and Ki used to locate the random position where the message bit

is to be embedded. These pixel indices are derived from the previous pixel indices, as

well as a subkey Ki, derived from the overall K. In general, the subkeys Ki are

different from K and from each other. The choosing of key is based Diffie-Hellman

key exchange algorithm which they used for encryption process that enables two users

to securely exchange a key and such algorithm depends for its effectiveness on the

difficulty of computing discrete logarithms (Stallings, 2006).

In the process of random pixel selection, two known numbers: a prime number „p‟

and an integer that is a primitive root of „p‟ are used. Firstly, in order to find a seed

for embedding a prime number „p‟ is searched that exceeds the key K. Then a

primitive root „a‟ of prime number is obtained that is a number whose powers

generate all the integers from 1 to p-1. That is, if „a‟ is a primitive root of the prime

number „p‟, then the numbers

a mod p, a2 mod p,..., a

p-1 mod p (7.4)

which are distinct and consist of the integers from 1 through p-1 in some permutation.

Each power of this primitive root to generate these integers is called the discrete

algorithm.

For any integer y and a primitive root „a‟ of prime number „p‟, a unique exponent i

can be found such that


selection

134

y ≡ ai (mod p), where 0 ≤ i ≤ (p-1) (7.5)

where the exponent i is referred to as the discrete logarithm of b for the base a, mod p.

This primitive root a, is then used to generate a set of random and distinct

numbers, yi = ai mod p, where i is the bit index of the secret message. Message bit „i‟

is then embedded into the LSB of the pixel yi and the order in which the secret

message bits are embedded is determined pseudo randomly. In this way it is ensured

that the bits of the secret message are inserted into distinct LSBs. The message is

encoded at pseudo-randomly chosen red planes. Thus embedding only in a single

plane will make small changes in the cover-image and such change is not visible and

detectable by the attacker. It could also make it more difficult for an attacker to figure

out the meaning of randomly extracted bits, incase an attacker is looking for evidence

of embedded information. By using a random number generator initialized with a

stego-key and its output is combined with the input data (secret message) which are

then embedded to a cover image. The usage of a stego-key is important, because the

security of a protection system should not be based on the secrecy of the algorithm

itself but also on choice of a secret key. In the random insertion method the random

location of the pixels depends on a stego key, whose size k should be in the range n <

k < I, where n is the size of message and I is the size of cover image. The main

property of a pseudorandom function is one-wayness. For a known input K, the

random sequence K1, K2,……….., Kn can easily be computed but it is very difficult

to reverse the process if K is unknown. The security of the Diffie-Hellman key

exchange lies in the fact that, while it is relatively easy to calculate exponentials

modulo a prime, it is very difficult to calculate discrete logarithms and for large

primes, the latter task is considered infeasible (Stallings, 2006). The purpose of using

prime number is that they occur in random.

Embedding Algorithm

i. Read character from text file that is to be hidden and convert the ASCII value

of the character into equivalent binary value into an 8 bit integer array.

ii. Read the RGB color image into which the message is to be embedded and

extract the red plane of the host image.


selection

135

iii. Initialize the key which gives the random position of the pixel‟s red plane to

be processed for embedding. The key produces random sequence of pixel

positions.

iv. Check the value of LSB (0 or 1) present in each pixel‟s red plane.

v. Pick up the message bit. If the message bit is zero and pixel bit is one then

decrement the pixel value. Else increment the pixel value. If message bit =

pixel bit, do nothing.

vi. Repeat step (iv) to (v) for the remaining random sequences (pixel indices).

vii. Add the green and blue plane with the modified red plane

viii. Write the above pixel to Stego-image file.

7.2.3 Technique to Retrieve secret message

To extract the message from the stego-image, the same stego-key is supplied to the

extraction algorithm which is used during the embedding process. This produces the

same sequence of random permutation or keystream based on the key. Because the

key integrates the random interval of pixels using a pseudo random number generator

(PRNG) to allocate the position of the cover image to embed the secret data. Since the

secret message is embedded in the cover-image‟s red plane thus in decoding only the

pixel‟s red plane is extracted and the correct key can take out the exact random

locations of the secret bits and thereby finding the correct seed (or starting value of

embedding), the value which is used to initialize the PRNG. The seed to initiate the

pseudo-random numbers is shared between the sender and the receiver and without

the correct seed the correct sequence of pixels will not be extracted. As the seed k is

known, ki can be reconstructed and therefore the entire sequence of pixel indices yi.

Even when knowing the image steganography algorithm, the receiver will only able to

extract a random string of data without knowledge of the order of the embedded

information. In other words, in extracting the message, the correct seed can trace the

embedding path and thus recovering the embedded bits. Therefore, the stego-key is

shared by both the sender and receiver during the secret communication.


selection

136

Extraction Algorithm

i. Open the Stego image, read the RGB color of each pixel.

ii. Extract the pixels red plane of the stego-image.

iii. Initialize the key that gives the position of the message bits in the random

pixels red plane where the secret bits are embedded.

iv. For decoding, select the pixels using the same pseudo-random sequence. As

the key knows the seed k, ki can be reconstructed and therefore the entire

sequence of pixel indices yi.

v. Read the LSB of the red plane of each pixel and put it directly in an array.

vi. Repeat step (v) for the entire sequence of pixel indices.

vii. Then content of the array converts into decimal value which is actually ASCII

value of hidden character.

7.3 Performance evaluation and Discussion

In this section, issues relating to the experimental approach are considered which is

used in the study to measure the impact of random embedding on the main aspects of

steganography. To evaluate the performance of the proposed scheme, several images

are used. In the first stage of the experiments, the proposed steganography method is

used to embed secret data in order to get a stego image. The secret data are embedded

in the random location of the red plane‟s LSBs in the cover-image which is

determined by a PRNG initiated by a stego-key. The experiments have been simulated

using the MATLAB R2009b program on Windows 7 platform using various

performance parameters. Here also, the PSNR is applied as performance measurement

criteria classified under the difference image distortion metrics and the MSE shows

variation between the cover- image and resultant (stego) image.

In the first case a color image baboon of size 512x512 is used as cover shown in figure

7.3 to embed secret messages of 18 K.B and 32 K.B and by using stego-key of 5000 and

15000. The messages are embedded into the red plane of the cover-image and the resulted

stego-image is as shown in figure 7.4 (a)-(d) respectively.


selection

137

Figure 7.3: Cover-image

(a) Stego-image with key=5000 (b) Stego-image with key=15000

Figure 7.4 (a) and (b) Stego-images formed by embedding 18 K.B (kilobyte) of secret

message using two different keys

(c) Stego-image with key=5000 (d) Stego-image with key=15000

Figure 7.4 (c) and (d) Stego-images formed by embedding 32 K.B (kilobyte) of secret

message using two different keys

From normal eyes perception, the result of the stego-images looks identical to the

cover-image as there appears no noticeable difference between them and the

complexity of the image is not disturbed as shown in figure 7.3 and 7.4 (a), (b), (c)


selection

138

and (d). This is because there are little changes of the pixel values and thus no

significant difference. Thus, in order to evaluate the efficiency of the proposed

steganography method, three dependent variables are considered: the steganographic

capacity, the quality of stego images (in PSNR and MSE) and the computational time.

These variables are the most commonly used aspects of image steganographic

systems. The computational time has been considered to investigate the impact of

embedding data using the pseudo random number generator for selecting the pixel and

embedding data on the processing time and has been measured for embedding and

extraction process. The results of the embedding process are listed in table 7.1.

Table 7.1: Results of embedding 18 K.B and 32 K.B of secret message into cover-

image baboon using two different stego-key

From table 7.1, it is observed that for all cases, PSNR is greater than 56 where the

message capacity ranges from 18 KB and 32 KB. On the other hand, the time

increases with the gradual increase in capacity. In addition, different key supplied also

generates different set of random numbers and thereby selecting different random

positions. If a bigger key size is chosen, the bigger the range of random numbers will

be generated, therefore the message bits will be scattered in a larger area. From table

7.1 it is seen that in embedding 18 Kb of secret data using key value 5000 the time to

embed and extract are 100.881718 and 96.411367 respectively whereas using key

value 15000 the time to embed and extract are 78.851781 and 75.386250 respectively,

thus the time computation decreases with increase in stego-key and same is the case

Cover-

Image

Message

Size

(in K.B)

key M.S.E P.S.N.R Time to

embed

(in seconds)

Time to

Extract

(in seconds)

baboon

18 kb

5000 0.0756 59.3438 100.881718 96.411367

15000 0.0755 59.3531 78.851781 75.386250

32kb

5000 0.1337 56.8699 543.146239 522.198800

15000 0.1339 56.8638 498.237917 479.118670


selection

139

for embedding 32 KB of secret data. Thus it can be stated that for larger key value the

time computation decreases.

As in the process of LSB embedding, a pseudo random number generator (PRNG)

which is seeded with a stego-key is used to select the random positions in the red

plane of the cover image by using a random interval method. The key produces a long

random sequence which is then used by the sender to generate the sequence of pixel

indices to get random location to embed the secret bits. Thus a wrong key will not

able to extract the random locations and thereby the original secret message cannot be

retrieved as shown in figure in figure 7.5.

Figure 7.5: Extracted Message with wrong key

This system has been developed to overcome a sequence-mapping problem when

using LSB which can be detected. However by using discrete logarithm calculation,

the problem of sequence-mapping can be solved. The work not only aims to preserve

the visual integrity of the image used for embedding but also the method should be

free from statistical attacks. So, distortion analysis of stego images is carried out by

analyzing statistical properties of the stego-images in comparison with the cover-

image. In the proposed method embedding of secret data is performed in the LSBs of

the cover-image. So, the method is also analyzed using statistical property by

visualizing the LSB bit-planes of cover-image and stego-images.


selection

140

Figure 7.6 Basic least significant bit of the cover image baboon (8-th bit)

(a) LSB of Stego-image with key=5000 (b) LSB of Stego-image with key=15000

Figure 7.7: (a) and (b) Basic least significant bit of the stego-images (8-th bit) of

baboon where the secret message of size 18 KB is embedded using two different keys

(c) LSB of Stego-image with key=5000 (d) LSB of Stego-image with key=15000

Figure 7.7: (c) and (d) Basic least significant bit of the stego-images (8-th bit) of

baboon where the secret message of size 32KB is embedded using two different keys


selection

141

The bit-plane of an image is a set of bits corresponding to a given bit position in each

of the binary numbers representing the image where the first bit plane contains the set

of the most significant bit, and the last part contains the least significant bit. In order

to demonstrate the effectiveness of the proposed method the least bit-plane of the

cover-image and the stego-images of baboon is presented in figure 7.6 and figure 7.7

(a)-(d) respectively. These are the bit-planes of the cover-image and stego-images of

figure 7.3 and figure 7.4 respectively. As the secret data is embedded into the LSB of

the red plane and thus present the overall LSB of the 8-th bit of figure 7.3 and figure

7.4 is presented where the data is embedded. Thus it can be seen that the difference in

the LSB bit-plane (8-th bit) where the secret bits are embedded cannot be traced as

there appears no major difference between the bit-plane of the cover-image and the

stego-images as seen in figure 7.6 and figure 7.7 (a)-(d). So, there appears no visible

difference between the bit-planes. This is because changing LSB either adds or

subtract 0 or 1. Also when the bit and the secret message and the LSB are same, then

no alternation of the pixels value takes place. Thus the proposed method preserves the

visual integrity of the image used for embedding but also the method is also free from

statistical and structural attacks.

In the second experiment, a comparative analysis of the proposed method with

Amirtharajan et al., (2013) scheme is conducted. Two well-known images Lena and

Rose each with size 250 × 250 are used as cover-images. The experiment is presented

to compare the impact of embedding 1.09 KB of secret message with different stego-

key values ranging from 2 to 205 on both the methods. The cover-image and the

resulting stego-images are presented in figure 7.8 and 7.9. For this case, the dependent

variables MSE and PSNR of red, green and blue plane are used for comparing.

http://en.wikipedia.org/wiki/Bit

http://en.wikipedia.org/wiki/Binary_number


selection

142

(a) cover-image (b) stego-image (c) stego-image

(d) stego-image (e) stego-image (f) stego-image

(g) stego-image (h) stego-image (i) stego-image

(j) stego-image (k) stego-image (l) stego-image

Figure 7.8: (a) Cover-image: Lena and (b)-(l) its resulting stego-images


selection

143

(a) cover-image (b) stego-image (c) stego-image

(d) stego-image (e) stego-image (f) stego-image

(g) stego-image (h) stego-image (i) stego-image

(j) stego-image (k) stego-image (l) stego-image

Figure 7.9: (a) Cover-image Rose and (b)-(l) its resulting stego-images


selection

144

Table 7.2: Results of comparison of the proposed method with Amirtharajan et al., (2013) method using Lena as cover-image and

message size of 1.09 K.B

Sample Key

values

Amirtharajan et al. (2013) Proposed Method

MSE PSNR MSE PSNR

Red Green Blue Red Green Blue Red Green Blue Red Green Blue

Key value=2 0.05152 0.4953 0.05004 61.01 61.18 61.14 0.0369 - - 62.46 - -

Key value=4 0.19328 0.194304 0.19712 55.27 55.25 55.18 0.0371 - - 62.44 - -

Key value=5 0.11340 0.115856 0.10875 57.58 57.49 57.77 0.0370 - - 62.44 - -

Key value=6 0.14828 0.140992 0.13337 56.42 56.64 56.88 0.0371 - - 62.44 - -

Key value=8 0.78848 0.8448 0.77721 49.16 48.86 49.22 0.0372 - - 62.43 - -

Key value=9 0.40867 0.390608 0.41651 52.02 52.21 51.93 0.0370 - - 62.45 - -

Key value=10 0.45043 0.4208 0.45363 51.59 51.89 51.56 0.0370 - - 62.45 - -

Key value=11 0.32524 0.295984 0.33129 53.01 53.41 52.93 0.0369 - - 62.46 - -

Key value=12 0.60876 0.516096 0.54758 50.28 51.00 50.75 0.0373 - - 62.41 - -

Key value=13 0.39052 0.374032 0.38801 52.21 52.40 52.24 0.0376 - - 62.37 - -

Key value=14 0.42124 0.411584 0.42278 51.88 51.99 51.87 0.0372 - - 62.43 - -

Key value=165 0.12148 0.108288 0.12128 57.28 57.78 57.29 0.0375 - - 62.38 - -

Key value=173 0.43398 0.370256 0.39676 51.75 52.44 52.14 0.0373 - - 62.41 - -

Key value=205 0.43862 0.369184 0.39641 51.71 52.46 52.15 0.0369 - - 62.45 - -


selection

145

Table 7.3: Results of comparison of the proposed method with Amirtharajan et al., (2013) method using Rose as cover-image

and message size of 1.09 K.B

Sample Key

values

Amirtharajan et al. (2013) Proposed Method

MSE PSNR MSE PSNR

Red Green Blue Red Green Blue Red Green Blue Red Green Blue

Key value=2 0.05050 0.05260 0.0498 61.09 60.92 61.15 0.0379 - - 62.35 - -

Key value=4 0.20120 0.20600 0.21555 55.09 54.99 54.79 0.0375 - - 62.39 - -

Key value=5 0.11620 0.11130 0.1131 57.48 57.66 57.59 0.0374 - - 62.40 - -

Key value=6 0.14140 0.12580 0.1424 56.62 57.13 56.59 0.0380 - - 62.33 - -

Key value=8 0.77920 0.85700 0.8386 49.21 48.30 48.89 0.0377 - - 62.37 - -

Key value=9 0.42408 0.42752 0.43091 51.86 51.82 51.78 0.0375 - - 62.39 - -

Key value=10 0.46750 0.46080 0.46144 51.43 51.49 51.49 0.0376 - - 62.38 - -

Key value=11 0.30530 0.30520 0.3025 53.28 53.28 53.32 0.0376 - - 62.38 - -

Key value=12 0.60210 0.58850 0.6097 50.33 50.43 50.28 0.0377 - - 62.36 - -

Key value=13 0.40430 0.37200 0.39712 52.06 52.42 52.14 0.0374 - - 62.40 - -

Key value=14 0.43100 0.39920 0.4055 51.78 52.12 52.05 0.0376 - - 62.38 - -

Key value=165 0.11228 0.11094 0.11161 57.63 57.68 57.65 0.0378 - - 62.36 - -

Key value=173 0.43810 0.39480 0.41724 51.71 52.17 51.93 0.0374 - - 62.39 - -

Key value=205 0.43448 0.39251 0.41582 51.75 52.19 51.94 0.0376 - - 62.38 - -


selection

146

The results in parts figure 7.8 (b) - (l) and 7.9 (b) - (l) are the stego-images of Lena and

Rose respectively which are produced by the proposed method. It can be seen that the

difference between the cover-image and stego-images cannot be perceived.

Figure 7.10: Comparison of PSNR values of the proposed method with Amirtharajan et

al., (2013) method with different key values for cover-image Lena

Figure 7.11: Comparison of PSNR values of the proposed method with Amirtharajan et

al., (2013) method with different key values for cover-image Rose


selection

147

In the second experiment, the effect of increasing the value of keys on the proposed

method and the Amirtharajan et al., (2013) method is evaluated. The results of this

experiment are shown in Table 7.2 and 7.3. In the proposed method as the embedding

takes place only in the red plane, so after performing the embedding operation green and

blue color planes are recombined into an RGB image to form the final stego-image. Thus

the MSE and PSNR values of the green and blue plane of the stego-images with respect

to cover-image does not change, as the pixel values of these planes remained the same

and no change occurrs in these planes. Thus in comparing the proposed method the

PSNR and MSE of the red plane shows variations. In the proposed method the PSNR

value of the red plane equals to the PSNR value of the cover-image. From the table 7.2

and 7.3 it is seen that the proposed method shows better PSNR value than the

Amirtharajan et al., (2013) in comparing the values of the red plane. It is also seen that

the PSNR value dynamically changes with the change in key value in Amirtharajan et al.,

(2013) scheme as the value of PSNR ranges from 50 dB to 61 dB whereas in the

proposed method the PSNR value are mostly static as the PSNR value are 62 dB for all

key values as shown in figure 7.10 and 7.11. The results show the difference between the

two methods, indicates that the quality and security of the stego-image is improved when

such randomization approach is utilized. A larger PSNR value indicates the fact that the

discrepancy between the cover image and the stego-image is more invisible to the human

eye. The results of the comparative evaluation are tabulated in table 7.2 and table 7.3 that

shows the PSNR values obtained demonstrate that the proposed scheme is superior to

Amirtharajan et al. (2013).

To evaluate further, two cover-images namely Desert and Tulips are used to embed secret

data for forming stego images. The secret messages used are of variable sizes ranging

from 4.81 K.B to 95.2 K.B that are embedded into desert and tulips. The resulting stego-

images of desert are plotted in figure 7.12 (b)-(f). For evaluation of the steganographic

technique several dependent variables are considered: the steganographic capacity, the

quality of stego images (PSNR and MSE), the size of stego images and secret messages,

the computational time etc. Therefore, all these variables are measured for the two test

images. The computational time has been considered to investigate the impact of random


selection

148

base embedding on the processing time and has been measured for both embedding and

extraction process. The results of embedding the secret messages into the cover images

are listed in Table 7.4.

(a)Cover-image (b) Stego-image

(c) Stego-image (d) Stego-image

(e) Stego-image (f) Stego-image

Figure 7.12: (a) Cover-image Desert and (b)-(f) the stego-images


selection

149

Statistical undetectability is one of the characters of a steganographic algorithm. The

statistical analysis compares the original image with the stego image based on histogram

(first order statistics) of images.

(a) Histogram of Cover-image (b) Histogram of Stego-image

(c) Histogram of Stego-image (d) Histogram of Stego-image

(e) Histogram of Stego-image (f) Histogram of Stego-image

Figure 7.13: (a) Histogram of cover-image Desert and (b)-(f) Histograms of the

corresponding stego-images


selection

150

Using histogram of cover image and stego-image the statistical property of the proposed

scheme is evaluated as shown in figure 7.13 (a) and figure (b)-(f) respectively. If the

distortion occurs then it can be suspected that the images may contain hidden data.

Comparing the histogram of the original channels, before and after modifying channels

can give a clear idea of the security; i.e. if change is minimal then the stego system is

considered secure. There seems no difference in the value of the pixel intensity in the

range 0 to 255 for the cover image and its stego image. The statistical change between the

original image and stego-image cannot be predicted as in figure 7.13 (a) and 7.13 (b) - (f)

respectively. The differences between the images before and after hiding the data cannot

be sensed through histograms of the RGB channels. It is seen, from different test runs at

different distributions between the three channels it is seen that they have particular

patterns in common.

Table 7.4: Shows the results of the proposed method using two cover-images Desert and

Tulips and secret message of variable sizes

Cover-

Image

Message

Size

(in K.B)

Stego-

image

Red

Plane

MSE

Red

Plane

PSNR

Overall

M.S.E

Overall

P.S.N.R

Time to

embed

(in

seconds)

Time to

Extract

(in

seconds)

Desert

4.81 stegdesert1 0.0074 69.44 0.0074 69.44 7.6925 1.20

15.5 stegdesert2 0.0237 64.38 0.0237 64.38 17.75 13.37

29.1 stegdesert3 0.0446 61.64 0.0446 61.64 58.40 66.86

64 stegdesert4 0.0942 58.39 0.0942 58.39 246.31 381.11

95.2 stegdesert5 0.1399 56.67 0.1399 56.67 711.50 940.28

Tulips

4.81 stegtulips1 0.0074 69.46 0.0074 69.46 8.17 1.08

15.5 stegtulips2 0.0236 64.40 0.0236 64.40 16.06 13.47

29.1 stegtulips3 0.0444 61.66 0.0444 61.66 58.70 65.62

64 stegtulips4 0.1019 58.05 0.1019 58.05 234.67 359.35

95.2 stegtulips5 0.1513 56.33 0.1513 56.33 693.32 911.18

In this evaluation process five sets of message ranging from 4.81 to 95.2 K.B in size are

embedded into cover-images Desert and Tulips as shown in table 7.4. The objective of


selection

151

this experiment is to show the effect of increasing the number of message size of the

proposed method on the PSNR and the computation time. It can seen that PSNR values

decreases gradually with the increase in message capacity and also the time computation

(embed and extract time) increases with increase in message. Thus it can be stated that

PSNR and computation time directly relates to the steganographic method. In this

evaluation the PSNR and the MSE value of the green and blue plane remains the same for

both cover-image and stego-image as the embedding takes place only in the red plane. So

in the table 7.4, the PSNR and MSE of red plane are plotted. The PSNR value of red

plane is same as the PSNR value of the overall image and so is uniformly distributed

among all the three channels as it uses all the planes. This improves the imperceptibility

and enhances the embedding capacity.

Figure 7.14: Shows the PSNR values when secret message of different sizes are

embedded into the cover-images desert and tulips

For both the cover-image, it is seen that the PSNR value gradually decrease from 69dB to

55 dB with the increase in message size from 4.81 K.B to 95.2 K.B as seen in figure 7.14.

So, increase in the amount of embedded data lowers PSNR value but still the PSNR

values are perfect and acceptable as embedding 95.2 K.B of secret data the PSNR values

for both the cover-image desert and tulips is 55 dB. The lower PSNR values owing to


selection

152

modifying more cover image pixels i.e. when more data are embedded. In other words,

the PSNR value of the cover images may be different due to the various sizes of secret

data. Certainly, it has higher PSNR values with lower modified rates.

Thus, to demonstrate the overall accomplished performance of the proposed approach in

evaluation process for hiding secret data in the cover-image, some experiments are

conducted using different cover-images of different resolutions to compare the proposed

approach with fixed and variable message size. Steganography message can be embedded

into digital images in ways that are imperceptible to the human eye. In other words, a

stego-image that is generated by the proposed algorithm is normal for human vision and

cannot be detected. The histogram behaviors of the images determine how large amount

of secret data can be embedded and still maintaining the statistical properties of the

resulting stego-images as seen in figure 7.13. By comparing tables 7.1 and 7.4, it is seen

that the actual PSNRs obtained are above the lower bounds. This serves as a good

predictor when any one image is to serve as a host for data embedding. PSNR values

falling below 30 dB indicate a low quality and a high quality stego-image should strive

for 40 dB and above (Cheddad et al. 2010). From various other reviews it is also found

that a color image should offer PSNR above 40 dB. From various experimental

evaluations it is found that the proposed method retrieves PSNR value above 56 dB

which means no significant degradation occurs.

Therefore, it is clearly seen from the experimental results that the performance of

proposed method is better. In addition to that, the advantage of proposed method is that

employment of stego key in embedding process provides better security. Results are

considered for color image of various dimensions using different stego-keys. It is also

seen in table 7.1 and table 7.4 that the computation time decreases by the number of key

increase which means that by increasing the number of key permutation, the Computation

time decreases and the System security increases. On the average case, the proposed

method embeds one bit in each pixel of red plane of a color image. The embedding

capacity can easily be enhanced by using two or three LSBs. In such case there will be a

decrease in PSNR value.


selection

153

In discussing the performance of the proposed method in terms of the stego-image quality

and the robustness against the detection attack. The proposed method does not select the

pixels LSBs of sequentially, instead it selects randomly based on key. By examining the

stego-images quality where the minimum and the maximum pixel-value alteration

happen, it is observed that the pixel alternation that takes place is negligible as seen in the

evaluation process. There appear no obvious suspicious artifacts on the resulting images

or bit-planes by statistical inspection as seen in figure 7.6 and 7.7.

In addition, the proposed scheme is secure against the well-known steganalysis like RS

detection attack or histogram analysis. The RS detection method proposed by Fridrich et

al. (2001) that targets LSB steganography. According to Fridrich et al. (2001), if more

and more LSBs are replaced with secret data, then the percentages regular and singular

values of two groups of pixels becomes more and more unequal. But from the overall

evaluation metrics the results of the method as shown indicate that the stego-images are

not suspicious, because the expected value of regular group of the cover-images to that of

the stego-images is close and so are the case singular groups as seen by verifying the bit-

planes (as in figure 7.6 and figure 7.7) and also the histogram (as in figure 7.13). This is

because the secret data are embedded only in the red plane whereas the green and blue

planes remain the same and thereby not altering much the pixel selected for data hiding.

This proves that the steganographic method is secure from the viewpoint of various

attacks. If the histogram does not display a smooth curve, the image is judged as a

suspicious image. In Figure 7.13, the histogram results of original image and stego-image

are plotted which displays a smooth curve. According to the overall analysis, it is

concluded that the proposed scheme is secure against the statistical and structural attacks.

7.4 Chapter Summary

In this chapter an approach of secure least significant bit (LSB) based steganography in

digital images that can override some statistical and structural measures of detecting

steganography. In this method, there is a provision of increasing the robustness by

spreading message bits randomly. Such approach is to overcome from the sequential


selection

154

LSB embedding where the message is encoded in the image file sequentially which

makes the detection easier. The primary characteristic of this steganographic technique is

that the data are embedded only in the red plane of the cover-image‟s pixel which is

determined by a PRNG initiated by a stego-key. By doing this, the method spreads the

secret data over the whole stego image randomly and the chance of getting the secret

message by the unintended receiver is reduced. Moreover the by selecting only the red

plane for embedding further enhances the security. The random number generator locates

the hiding positions in the red plane of the cover image by using a random interval

method. Here the key is used not only as password but also as a random position locater

in the cover-image and which becomes significantly difficult for the attacker to figure out

the target pixels. Since embedding operated only in red plane thus in decoding only the

pixel‟s red plane is selected and the same stego-key is used. In examining the proposed

technique, it is seen from the overall evaluation metrics the performance of the proposed

method indicate good results and is found to be robust against many attacks. The method

proposed has been empirically validated as to indicate its utility and value.

CHAPTER 7 Steganography based on Random pixel selectionshodhganga.inflibnet.ac.in/bitstream/10603/97221/9/th-1816_ch7.pdf · image based steganography is selecting of technique that

Documents

CHAPTER 7 Steganography based on Random pixel selectionshodhganga.inflibnet.ac.in/bitstream/10603/97221/9/th-1816_ch7.pdf · image based steganography is selecting of technique that