A Spatial Domain Image Steganography Technique Based on Matrix Embedding and Huffman Encoding

P.Nithyanandam, T.Ravichandran, N.M.Santron & E.Priyadarshini
International Journal of Computer Science and Security (IJCSS), Volume (5) : Issue (5) : 2011 456
A Spatial Domain Image Steganography Technique Based on
Matrix Embedding and Huffman Encoding
P.Nithyanandam nithyanandamp@ssn.edu.in
Department of Computer Application
SSN College of Engineering,
Anna University of Technology, Chennai
Kanchipuram Dt, Tamilnadu , 603110,India
T.Ravichandran dr.t.ravichandran@gmail.com
Principal
Hindustan Institute of Technology,
Anna University of Technology, Coimbatore
Coimbatore Dt,Tamilnadu, 641032,India
N.M.Santron nmsantron@gmail.com
III Year M.C.A.
E.Priyadharshini indrapriyadharshini.e@gmail.com
III Year M.C.A.
Abstract
This paper presents an algorithm in spatial domain which gives less distortion to the cover image
during embedding process. Minimizing embedding impact and maximizing embedding capacity
are the key factors of any steganography algorithm. Peak Signal to Noise Ratio (PSNR) is the
familiar metric used in discriminating the distorted image (stego image) and cover image. Here
matrix embedding technique is chosen to embed the secret image which is initially Huffman
encoded. The Huffman encoded image is overlaid on the selected bits of all the channels of
pixels of cover image through matrix embedding. As a result, the stego image is constructed with
very less distortion when compared to the cover image ends up with higher PSNR value. A secret
image which cannot be embedded in a normal LSB embedding technique can be overlaid in this
proposed technique since the secret image is Huffman encoded. Experimental results for
standard cover images, which obtained higher PSNR value during the operation is shown in this
paper.
Keywords: Steganography, Imperceptibility, Payload, Stego Image, Least Significant Bit (LSB),
Huffman Encoding, Matrix Embedding, Peak Signal to Noise Ratio (PSNR), Mean Square Error
(MSE) and Discrete Wavelet Transformation (DWT).
1. INTRODUCTION
Steganography is the art of secret communication. It has apparent difference with cryptography;
because cryptography hides information content whereas steganography hides information
existence. Steganography is broadly classified in to spatial and frequency domain technique.
Least Significant Bit (LSB) replacement, LSB matching, Matrix embedding and Pixel value

differencing are some of the spatial domain techniques. Frequency domain techniques include
Outguess, F5, JP Hide and Seek. Fundamentally, a steganography algorithm or embedding
function can influence the cover work in three different ways, namely cover lookup, cover
synthesis and cover modification. Naturally, changes of larger scale will be more obvious than
changes of smaller scale. As a result, most steganographic schemes try to minimize the distortion
on cover work. The location of changes is controlled by the selection rule [1]. There are three
types of rule namely sequential, random and adaptive.
The primary goal of steganography is to design embedding function that should be statistically
undetectable and capable of communicating large payloads. There exists a tradeoff between
embedding capacity and proportion of distortion. There are many algorithms evolving to
accomplish steganography goal in both spatial and frequency domain. Minimizing the embedding
impact while constructing a stego image could be one of the ways; this may thwart in applying
statistical analysis over a stego image. The notion of this paper is to apply one such embedding
technique and to produce a less distorted cover image. Supporting a higher payload on a cover
image depends upon embedding technique; but it also can be viewed in another direction of
compressing the payload before overlaying. A lossless Huffman [2] [3] [4] [5] compression prior to
overlaying results in fewer distortion in the cover image.
Cachin’s [1] description of steganography security calls for the Kullback-Leibler distance which
says, the probability distance between the cover and stego work to be as little as possible. In our
technique it is achieved by minimizing the distortion between the cover and stego work. This will
make it harder for the warden to detect embedding. The embedding procedure can encode the
message bits in many ways. For example in LSB embedding the LSB is replaced to match the
secret message bits. On average, one can embed, 2 bits per embedding change. It can be
substantially improved if we adopt a clever embedding scheme. In particular, if the payload is
shorter than the embedding capacity, one can influence the location of changes to encode more
bits per change. Let us take a look at the following simple example. Say, we have a group of
three pixels with gray scale values x1, x2 and x3. We wish to embed 2 message bits, b1 and b2.
It seems that a practical approach might be to simply replace b1 with x1 and b2 with x2 (i.e.)
replacing the LSB of the pixels to match the corresponding message bits. Assuming the 2 bits are
0 or 1 with equal probability, the expected number of changes to the whole group of pixels to
embed both bits is 1. Therefore, we embed at embedding efficiency of 2 or 2 bits per change.
However, it can be improved. Let us encode b1 = LSB (x1) XOR LSB (x2) and b2 = LSB (x2)
XOR LSB (x3). If the values of the cover work satisfy both equations with equality, no embedding
changes are required. If the first one is satisfied but not the second one, simply flip the LSB of x3.
If the second one is satisfied but not the first one, flip the LSB of x1. If neither one is satisfied, flip
LSB of x2. Because all four cases are equally likely with probability 1/4, the expected number of
changes is 3/4, which is less than what we had earlier. This embedding technique is called matrix
embedding [1] which is further extended and used in the proposed method.
Huffman compression is a variable length coding whose performance depends on the input
image bit stream. The compression is directly proportional to smoothness of the image. Higher
the smoothness and higher the redundancy will give good compression. Subjective and objective
measures [6] are the two techniques existing to test the distortion of the processed image.
Subjective measure is not reliable because human vision is a metric in assessing the distortion of
the stego objects. Human vision may vary from person to person; hence this approach is not
suitable. In objective measure, the mean square error (MSE) represents the cumulative squared
error between the stego image and cover image. A lower figure of MSE conveys lower error/
distortion between the cover and stego image.

∑∑− −
−=
m
i
n
j
ijij BA
nm
MSE
1 1
2
)(
*
1
The equation of MSE to assess the stego and cover object is given by:
...........[1]
Whereas Aij represents pixel in the cover image and Bij represents pixel in the stego image; m, n
represents the height and width of the image respectively. It is measured in constant and the unit
is decibel (dB).
Peak Signal to Noise Ratio (PSNR) is a metric which calculate the distortion in decibels, between
two images. Higher the PSNR indicates a better reconstructed or stego image. The PSNR is
represented by the following equation:
MSE
Max
PSNR
2
10
)(
log*10= ...………..[2]
Where max denote maximum intensity of grayscale (255).PSNR is measured in decibels (dB).
2. RELATED WORK
Chang, C.C et al., [7] has proposed an image steganography technique which offer high
embedding capacity and bring less distortion to the stego image. The embedding process embed
bits of secret bit stream on the stego image pixels. Instead of replacing the LSB of every pixel,
this method replaces the pixel intensity with similar value. The range of modifiable pixel value is
higher in edge areas than smooth areas to maintain good perceptual excellence. Various bit
embedding methods are followed; which are decided by the correlation between the actual pixel
and the neighboring pixels. The neighboring pixels may be a pixel left, right, top or bottom to the
actual pixels. The different schemes are two sided, three sided and four sided one. Two sided
scheme take upper and left pixels, three side scheme take upper, left and right whereas four
sided take upper, left, and right and bottom pixels. The embedding capacity and PSNR are
inversely proportional to the sides taken into account.
Po-Yueh Chen et al., [8] proposed an image steganography scheme which fixes the limitation of
steganography technique proposed in [7]. The limitation of [7] is falling of boundary problem
which means the pixel which is located for embedding will become unused; since it exceeds the
maximum intensity level which is greater than 255 (maximum gray scale intensity). Fewer bits are
added even on such pixels which improve the embedding capacity without compromising PSNR
in this technique.
A. Nag et al., [9] proposed a stenographic technique which is based on wavelet transformation on
the images. Discrete Wavelet Transformation (DWT) converts the spatial domain of cover image
into frequency domain. Huffman compression is applied for the stream of secret bits before
overlaying them on the cover image. A high PSNR and very high embedding capacity is
achieved.
R.Amirtharajan et al., [10] proposed a stenographic technique which is based on LSB
replacement technique. Varying lengths of secret bits get embedded in every pixel. In method1
green and blue are embedding channels keeping red channel as indicator channel. In method2
an option is provided for choosing the indicator channel among the three channels. Once chosen,
the remaining two channel act as embedding channel. In method3 the indicator channel is chosen
by rotation scheme across all the pixels. In the first pixel red channel is indicator; green channel is
the indicator in second pixel and in third channel blue act as indicator. Once indicator is finalized
the remaining two channels will be used for embedding. This scheme is repeated for the
consecutive pixels. The MSE and PSNR is calculated for all channel and the average number of
bits get embedded in every pixel is shown in their results.

The rest of the paper is organized as follows. Section III discusses the proposed steganography
technique. In Section IV experimental results are exhibited and discussed. Finally the conclusion
and future direction are provided for the proposed work.
3. PROPOSED METHOD
3.1. System Architecture
Fig.1 shows the overall system architecture on which the proposed study stands on. The secret
image pixel values are Huffman compressed which comprises of Huffman encodings and
Huffman table. The size of Huffman table and Huffman encodings are measured in a 32 bit
quantity each. These 64 bits are recorded across the last 64 byte’s LSB of the stego image. Both
the Huffman encodings and Huffman table binary content are embedded in the LSB of every byte
using LSB replacement or Matrix embedding technique. The binary content of Huffman table is
followed by Huffman encodings. The starting and the ending point of the corresponding binary
component i.e. Huffman encodings or Huffman table is identified through the processed individual
32 bits entry stored at the end of the stego image. In the case of the secret image being
sufficiently large, the stego image LSB may be fully utilized. Always, the last 64 byte is reserved
for storing the size of Huffman table and Huffman encodings.
FIGURE 1: Stego Image Architecture
3.2. Huffman Compression on Image
The intensity of the pixels across the image is spatially correlated [3]. Information is pointlessly
repeated in the representation of the correlated pixels. These repetitive pixels should also be
represented by fixed number of bits in unencoded Huffman format. Actually these values are the
best source for exploiting compression. A very frequent occurrence intensity value can be
represented by variable numbers of bits (i.e. shorter bits) in contrast to the fixed number of bits for
representing the pixel intensity used in unencoded Huffman technique. This is the core concept of
Huffman encoding technique. The secret image is Huffman encoded prior to embedding process.
3.3. Extended Matrix Embedding
An extended matrix embedding technique is used in proposed method. Generally (1, n, k) matrix
embedding [11] mechanism is used; which denotes k secret bits are embedded in n cover bits
with at most 1 change. Here using three Least Significant Bits of RGB channel 2 bits of secret
bits might be embedded with at most one change, which is typically (1,3,2) in the above case.
Here n is 2k
-1.
It can be further expanded by considering; more secret bits can be embedded in a single go with
at most 1 change. For example if k is 3, then n is 2
k
-1. K secret bit should be embedded in 2
k
-1
LSB of every byte: Huffman Encodings embedded
LSB of every byte: Huffman Table embedded
Non modified part: may be utilized in the case of
secret image size is large
32- bit (length of
Huffman table)
32- bit (length of
Huffman
encodings)

cover bit with at most 1 change. It is denoted by (1,7,3), where 1 represent number of changes
allowed,7 represent number of cover bit involved in the operation and 3 represent number of
secret bit to be embedded. Now the cover bit selection and embedding mechanism to be
designed in such a way that, k secret bits should be embedded in n cover bits with at most 1
change.
1) Cover bit Selection: Two types of cover bit selection are attempted in the above proposed
technique and the results are shown for both the types.
Method1: In this method the LSB of every byte is chosen as cover bit. 7 bits of data are required
to embed a 3 bit secret data. Those 7 bits are collected from seven consecutive bytes of the
image. All 7 bytes’ LSB is serving as cover bit.
Method2: In this method to collect 7 cover bit for the operation, on every pixel last two bits of red
channel, last three bits of green channel and last two bits of blue channel are taken.
2) Secret bit Embedding: In order to embed and extract the 3 secret bit in the 7 cover bit with
atmost 1 change, a reversible embedding and extraction algorithm should be designed. Equation
3 shown below will be used to meet the above goal. Assume b1,b2,b3 are the secret bits,
x1,x2,x3,x4,x5,x6,x7 are cover bits. The cover bits are adjusted according to the secret bits b1, b2
and b3 with atmost 1 change i.e. only one change is permitted out of all the 7 cover bits. At the
same time the secret bit should be mapped inside the cover bit. The following equation is used in
both embedding and extraction process.
. ....................[3]
The above 3 expression in equation 3 is operated to check the coincidence of secret bit against
cover bit. An exclusive OR operation is performed on the cover bit; if all the three expression is
satisfied no adjustment is required on the cover bit. Sometimes the cover bit by itself, is suitable
to fit the secret data. If any or more than one of the expressions in equation 3 is not satisfied then
modification on the cover bit is followed according to Table 1. This slight modification on the cover
bit enable the secret bit to be mapped on the cover bit with at most only one change. Since the
cover bit are adjusted according to the secret bit; during extraction the same equation can be
used in recovering the secret bit from the cover bit.
Secret
bit
Positions
not
matched
(b1,b2,b3)
1 2 3 1,2 2,3 1,3 1,2,3
Cover bit
to be
inverted
x1 x2 x3 x4 x5 x6 x7
TABLE 1: Embedding/ Extraction Reference
76533
75422
76411
xxxxb
xxxxb
xxxxb
⊕⊕⊕=
⊕⊕⊕=
⊕⊕⊕=

Huffman coding technique is used in the proposed method to securely and imperceptibly hide the
secret image in the cover image. The Huffman encoded bit stream and Huffman table bit stream
is embedded in the cover image pixel either by method1 or method2 through Matrix embedding
technique. Cover bit selection will differ in method1 and method2 whereas embedding process
remain same.
3.4. Hashing
Finally to attain the integrity of the stego image; the secret image is hashed prior to embedding.
This hash code should be send as a supplementary component in addition to stego image. In the
receiving end, the retrieved secret image is hashed to cross check against the hash code
received. If both the hash codes are same, it conveys no intruder has modified the stego image.
3.5. Embedding Process
Fig. 2a shows the embedding process carried on the sender side. The Hash code of secret image
and stego images are sent to receiver.
The steps carried on the sender side are given below:
Step 1: Hash the secret image.
Step 2: The Secret image is converted into a Numerical matrix which contains the RGB value or
intensity of each pixel.
Step 3: Apply Huffman encoding for the output obtained from Step 2 which results in Huffman
table and Huffman encoded secret image bit streams.
Step 4: Group the above obtained binary bit stream (Huffman table and Huffman encoded) in
chunk of three bits.
Step 5: M1: Method1:- Each color image pixel is represented by 3 bytes (RGB). Collect 7
consecutive bytes from the image. All 7 bytes’ LSB is serving as cover bit.
Step 6: M1: Method1:- Using equation 3 adjusts the 7 bytes LSB to match the three secret bit
chunk obtained in Step 4. (OR)
Step 5: M2: Method2:- Each color image pixel is represented by 3 bytes (RGB). In this method
to collect 7 cover bit for the operation, on every pixel LSB and LSB -1 from Red
channel, LSB, LSB -1 and LSB -2 from Green channel, LSB and LSB -1 from Blue
channel; a total of 7 bits are chosen as cover bit.
Step 6: M2: Method2:- Using equation 3 adjusts the above 7 bits to match the three bit chunk
obtained in Step 4.
Step 7: Repeat Step5 and Step6 until all the 3 secret bit chunks are mapped over the cover
image pixels moving from left to right and top to bottom of the cover image.
Step 8: Send the Hash Code and stego image obtained from Step 7 to the receiver.
3.6. Extraction Process
Fig. 2b shows the extraction process carried on the receiver side. Upon receiving the stego
image, and the Hash code, receiver should extract the Huffman table, Huffman encoded bit
streams, and secret image dimension from the stego image.
The steps carried on the receiver side are given below:
Step 1: Apply the relevant bit collection on the stego image pixel depends on the method
(method1/method2); the secret bit is embedded in the cover image as explained in
embedding process.
Step 2: Size of secret image, Huffman Table and Huffman symbols are retrieved.
Step 3: The Binary Huffman table is then converted to the actual format that can be accepted
by the Huffman decoding.
Step 4: The Huffman table and Huffman encodings obtained in Step 2 are used in Huffman
decoding process. As a result RGB/intensity value, for every pixel of secret image is
obtained.
Step 5: Finally, the image is constructed using all the pixels which is computed in Step 4 will
reveal the secret image.
Step 6: To ensure the stego image integrity, the received hash code is compared against the
Hash code of constructed secret image. If both are equal, cover image is free from

FIGURE 2a: Embedding Process
Binary
Conversion
Cover Image
Huffman
Encoding
Secret Image
Huffman
Encoded
Data
Huffman
Table
Matrix
embedding
Stego Image
Hashing
Hash file
intruder attack.
The intermediate results obtained in every stage of embedding and extraction process are
redirected to a text file may be assumed for better understanding of the proposed method
wherever required.
4. EXPERIMENTAL RESULTS
Java 2.0 and MATLAB 7.6 are the programming tools used to implement the proposed method.
PSNR, Embedding Capacity and Mean Square Error are the three metrics taken here to
consolidate the strength of proposed method. PSNR result is shown separately for all the
channels. Two tables are used to present the performance of both the methods. The same cover
image of size 256 X 256 is used in both the methods. The cover image and secret image taken
here for experimentation is 24 bit color depth bmp (Bit Map Format) image.
A secret image Cameraman Fig. 3 of various sizes is embedded in the RGB cover images like
Lena, Airplane, Baboon and Boat each of size 256 x 256. Fig. 4-7 shows the cover images,
obtained stego images and histogram arrived in method1 and method2 of matrix embedding
technique. Table2 and Table3 show the experimental results of method1 and method2
respectively. The PSNR and MSE arrived using the proposed method shows that the distortion
occurred in stego image are very less. In method1 secret image of different sizes such as 85x85,
90x90 and 95x95 with 24 bit depth are embedded. The maximum capacity that the cover image
can hold is 216,600 bits which is 26.5KB. The embedding capacity is 14% of the cover image
using method1. The average PSNR and mean in method1 for 95x95 secret image is 58 and 0.12
respectively.
In method2, since the 7 cover bits are collected on a single pixel, the embedding capacity of the
same cover image is better than method1. In method2, the same secret image Cameraman Fig. 3
of different size such as 85x85, 90x90, 95x95, 140x140, 150x150, and 155x155. In method2 a
higher capacity is achieved but PSNR and mean is compromised. The maximum capacity that the
Stego Image
Matrix extraction
FIGURE 2b: Extraction Process
Huffman encoded
binary Stream
Huffman Table
Huffman
Decoding
Size of secret
message and
Huffman Table
Secret Image
Decimal
Conversion
Hash file
Hashing Hash file
C
O
M
P
A
R
E

cover image can hold is 576,600 bits which is 70.38KB. The embedding capacity is 37% of the
cover image using method2. The average PSNR and mean in method2 for 155x155 secret image
is 50 and 0.6 respectively. The PSNR and mean has declined with an enhanced capacity; but still
PSNR value with more than 40 is acceptable.
Cover Image of
size 256 X 256
Red
Channel
Green
Channel
Blue
Channel
PSNR MSE PSNR MSE PSNR MSE
Lena
85 x 85 57.94 0.1044 57.90 0.1052 57.89 0.1057
90 x 90 57.63 0.1120 57.45 0.1169 57.51 0.1151
95x 95 57.18 0.1243 57.22 0.1232 57.11 0.1263
Airplane
85 x 85 57.94 0.1044 57.89 0.1057 57.82 0.1072
90 x 90 57.51 0.1151 57.61 0.1125 57.46 0.1164
95x 95 57.23 0.1227 57.12 0.1259 57.19 0.1242
Baboon
85 x 85 57.87 0.1061 57.93 0.1046 57.87 0.1060
90 x 90 57.54 0.1145 57.55 0.1141 57.49 0.1156
95x 95 57.15 0.1252 57.22 0.1232 57.17 0.1246
Boat
85 x 85 57.95 0.1040 57.88 0.1059 57.82 0.1073
90 x 90 57.56 0.1139 57.55 0.1141 57.46 0.1167
95x 95 57.15 0.1251 57.14 0.1254 57.14 0.1255
TABLE 2: 7 COVER BIT ON 7 BYTE (METHOD 1)
Cover Image of
size 256 X 256
Red
Channel
Green
Channel
Blue
Channel
PSNR MSE PSNR MSE PSNR MSE
Lena
85 x 85 54.55 0.2277 48.40 0.9393 54.55 0.2278
90 x 90 54.15 0.2497 47.96 1.0397 54.30 0.2411
95x 95 53.81 0.2700 47.72 1.0969 53.87 0.2665
140x140 51.02 0.5131 44.77 2.1646 50.97 0.5200
150x150 50.50 0.5783 44.17 2.4874 50.37 0.5959
155x155 50.25 0.6131 43.96 2.6100 50.22 0.6169
Airplane
85 x 85 54.56 0.2275 48.40 0.9379 54.57 0.2266
90 x 90 54.28 0.2423 47.94 1.0445 54.20 0.2469
95x 95 53.81 0.2700 47.57 1.1366 53.89 0.2652
140x140 50.98 0.5180 44.80 2.1505 51.02 0.5134
150x150 50.45 0.5856 44.16 2.4921 50.51 0.5781
155x155 50.23 0.6155 44.04 2.5631 50.15 0.6275
Baboon
85 x 85 54.57 0.2267 48.34 0.9527 54.64 0.2229
90 x 90 54.33 0.2397 47.97 1.0374 54.12 0.2517
95x 95 53.89 0.2652 47.70 1.1028 53.75 0.2740
140x140 51.01 0.5145 44.73 2.1881 50.97 0.5190
150x150 50.47 0.5824 44.24 2.4460 50.43 0.5882
155x155 50.21 0.6191 43.98 2.6004 50.16 0.6266
Boat
85 x 85 54.61 0.2248 48.27 0.9673 54.56 0.2272
90 x 90 54.24 0.2448 47.89 1.0569 54.27 0.2427
95x 95 53.86 0.2673 47.72 1.0984 53.81 0.2699
140x140 51.04 0.5110 44.76 2.1710 51.01 0.5147
150x150 50.43 0.5882 44.25 2.4384 50.48 0.5815
155x155 50.30 0.6064 43.93 2.6298 50.23 0.6160
TABLE 3: 7 COVER BIT ON 1 PIXEL – 2, 3,2 (METHOD 2 )
The proposed method’s hiding capacity depends upon the Huffman encoding output. The
Huffman encoded result of a secret image (Huffman encoded bit stream and Huffman Table) size
should be lesser than the total number of LSB spot available in the cover image. The last 64 pixel

in cover image is reserved for storing the technical details which will be used in the receiver side
to extract the secret image from the stego image. This 64 pixel (64x3=192 bytes) should be
excluded while computing the hiding capacity of cover image. Image of any size/richness can be
hidden through our proposed method, provided it meets the above said condition. Integrity of the
stego image is verified by crosschecking the hash code received against the constructed secret
image hash code. If both hash code are same, it conveys no intruder modified the stego image.
4.1. Discussion
In method2 the PSNR of green channel is less, compared to the other two channels. It is due to
the reason that the cover bits are selected in the same pixel in this order (2, 3, and 2). Two bits
from red channel, three bits from green channel and two bits from blue channel are taken. Out of
7 bits, 3 bits are taken from green channel; hence this channel is highly vulnerable to distortion.
So, as a result the PSNR of green channel has declined in method2.
We quite often found that a secret image which is richer and whose dimension is lesser than
Cameraman,(shown in Fig. 3) say 100 X 100 cannot be embedded in this 256 X 256 cover image
shown in figure 4. In contrast, a secret image which is not richer and whose dimension is higher
than 100 X 100 can be embedded in the cover image. This makes us to finalize that the
embedding capacity of our proposed technique depends on Huffman encoding. Any image,
whose Huffman compression is less, fits in the cover image irrespective of its size and richness.
The embedding capacity of the cover image can be improved further, if a pixel adjustment
process technique is adapted. The number of bits get embedded in the proposed technique is just
3 bit per pixel in method1 or 3 bit using LSB’s of seven consecutive bytes in method2. Pixel
adjustment process technique is just substituting the intensity of the every cover pixel with an
equivalent resembling pixels. This could exploit the cover pixels in embedding greater than 3 bits
(9 bits/pixel). But, it will be on the cost of compromising a little bit distortion gets introduced on the
cover image.
To discuss on security side, the proposed technique is robust enough; because extracting a data
without knowing the architecture of the proposed technique is difficult, moreover data is Huffman
encoded. Stego image integrity is validated through hashing which give confidence to the
receiver. Thus, the privacy and security issues are addressed in this proposed technique to a
reasonable extent.
CONCLUSION
We had proposed an image steganography algorithm which brings a better PSNR and MSE. The
experimental results show that distortion between cover and stego image is minimum. Capacity
improvement and distortion reduction has been addressed in this proposed technique. In the
proposed method, the embedding capacity of the cover image is increased which results in slight
decline in both PSNR and MSE parameters. The veracity of the stego image is verified and then
progressed for their usage on receiver side. The proposed technique is not robust against any
geometrical distortion such as rotation, translation, scaling, cropping etc., induced on the stego
image. Improving this parameter is still under research and not matured yet.
FUTURE WORK
The proposed algorithm should be customized to support embedding in the frequency domain. It
should be enhanced to withstand geometrical distortion induced on the image.

FIGURE 3: Cameraman
FIGURE 4a: Lena Cover FIGURE 4b: Red Channel FIGURE 4d: Blue ChannelFIGURE 4c: Green Channel
FIGURE 4e: Lena Stego M1 FIGURE 4f: Red Channel FIGURE 4h: Blue ChannelFIGURE 4g: Green Channel
FIGURE 4l: Blue ChannelFIGURE 4i: Lena Stego M2 FIGURE 4j: Red Channel FIGURE 4k: Green Channel

FIGURE 5c: Green ChannelFIGURE 5a: Airplane Cover FIGURE 5b: Red Channel FIGURE 5d: Blue Channel
FIGURE 5i: Airplane Stego M2 FIGURE 5j: Red Channel FIGURE 5l: Blue ChannelFIGURE 5k: Green Channel
FIGURE 5h: Blue ChannelFIGURE 5e: Airplane Stego M1 FIGURE 5f: Red Channel FIGURE 5g: Green Channel
FIGURE 6a: Baboon Cover FIGURE 6b: Red Channel FIGURE 6d: Blue ChannelFIGURE 6c: Green Channel

FIGURE 6e: Baboon Stego M1 FIGURE 6f: Red Channel FIGURE 6h: Blue ChannelFIGURE 6g: Green Channel
FIGURE 6i: Baboon Stego M2 FIGURE 6j: Red Channel FIGURE 6l: Blue ChannelFIGURE 6k: Green Channel
FIGURE 7d: Blue ChannelFIGURE 7a: Boat Cover FIGURE 7b: Red Channel FIGURE 7c: Green Channel
FIGURE 7h: Blue ChannelFIGURE 7e: Boat Stego M1 FIGURE 7f: Red Channel FIGURE 7g: Green Channel
FIGURE 7l: Blue ChannelFIGURE 7i: Boat Stego M2 FIGURE 7j: Red Channel FIGURE 7k: Green Channel

REFERENCES
[1] Injemar J. Cox, Matthew L. Miller, Jeffrey A. Bloom, Jessica Fridrich, Ton Kalker. Digital
Watermarking and Steganography. Morgan Kaufmann, Second Edition,2008.
[2] Professor Luca Trevisan, 2001, "Lecture notes on Intro. To CS Theory." Online.
Available: http:// ww.cs.berkeley.edu/~luca/cs170/notes/lecture15.ps
[3] Rafael C. Gonzalez Richard E. Woods. Digital Image Processing. ,PHI, Third Edition,
2008.
[4] Alasdair McAndrew. Introduction to Digital Image Processing with MATLAB, CENGAGE
Learning, 2004
[5] [Online].Available : http://www.binaryessence.com
[6] Ali K. Hmood, Z. M. Kasirun, Hamid A. Jalab,Gaz Mahabubul Alam, A. A. Zaidan, B. B.
Zaidan. “On the accuracy of hiding information metrics: Counterfeit protection for
education and important certificates.” International Journal of the Physical Sciences,
Volume. 5, Issue 7, pp. 1054-1062, August,2010.
[7] Chang, C.C and Tseng, H.W. 2004. “A Steganographic method for digital images using
side match.” Pattern Recognition Letters, 25: 1431 – 1437, June 2004.
[8] Po-Yueh Chen,Wei-En Wu. “A Modified Side Match Scheme for Image Steganography” ,
International Journal of Applied Science and Engineering, Volume 7, Issue 1, pp. 53-60,
October 2009.
[9] A. Nag, S. Biwa’s, D. Sarkar, P.P. Sarkar. “A Novel Technique for Image Steganography
Based on DWT and Huffman Encoding” , International Journal of Computer Science and
Security, Volume 4, Issue 6, pp. 561-570, Feb. 2011.
[10] R.Amirtharajan, Sandeep Kumar Beher, Motamarri Abhilash Swarup, Mohamed
Ashfaaq K and John Bosco Balaguru Rayappan. “Colour Guided Colour Image
Steganography”, Universal Journal of Computer Science and Engineering Technology,
Volume 1, pp.16 – 23, Oct . 2010.
[11] Santosh Arjun, Atul Negi, Chaitanya Kranti, and Divya Keerthi. “An Approach to Adaptive
Steganography Based on Matrix Embedding” TENCON 2007 - 2007 IEEE Region 10 ,
Volume 1,pp.1-4, Oct . 2007.

A Spatial Domain Image Steganography Technique Based on Matrix Embedding and Huffman Encoding

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to A Spatial Domain Image Steganography Technique Based on Matrix Embedding and Huffman Encoding (20)

Recently uploaded (20)

A Spatial Domain Image Steganography Technique Based on Matrix Embedding and Huffman Encoding