Special Issue - 2016 International Journal of Engineering Research & Technology (IJERT)

ISSN: 2278-0181
ICACT - 2016 Conference Proceedings

Multiple Image Steganography using LSB-DCT

Technique
Devadath C Prabhu S Ramya Nivedha Ayush Kumar
Department of CS&E, Department of CS&E, Department of CS&E,
PESIT South Campus, PESIT South Campus, PESIT South Campus,
Bengaluru, India Bengaluru, India Bengaluru, India

Annapurna D
Sajeevan K Professor,
Associate Professor, Department of CS&E,
Department of CS&E, PESIT South Campus,
PESIT South Campus, Bengaluru, India
Bengaluru, India

Abstract— Steganography is an art of secret

communication. The very existence of the data itself is hidden in
steganography. Any little variation in pixels of the images has a
low probability of being detected by the human eyes. Thus, data
can be hidden in pixels of images. In this paper, a LSB-DCT
based steganographic technique for hiding multiple images in a
cover image is proposed. The data is embedded by altering least Fig 1. Steganographic system
significant bits of quantized DCT coefficients of cover image.
Hiding multiple images is achieved by embedding only higher There are different Steganographic techniques [4] each
bits of data in cover image. Experiments shows that with their own advantages and disadvantages [5], [6], [7]. In
imperceptibility is high for the stego image as well as PSNR this paper, the secret image is hidden in cover image using
values obtained for both stego image and extracted images are combination of LSB and DCT techniques. The cover image is
good. transformed into frequency domain by applying DCT-II.
Then pixel bits of secret image is hidden in LSB of quantized
Keywords—Steganography, LSB, DCT, Cover Image, Stego DCT co-efficients of cover image and is transformed back to
Image, PSNR spatial domain to construct the stego image. Few factors
I. INTRODUCTION considered while designing the system are imperceptibility,
payload capacity, PSNR and MSE which are discussed in
Secrecy of information is one of the most important factor detail in Section IV. The process of detecting the hidden
in communication. There are two main methods by which messages without knowing the key and the algorithm used is
secrecy of information can be obtained – Cryptography and Steganalysis [16].
Steganography. Cryptography is the process of using codes to The remainder of the paper is structured as follows.
ensure secrecy of information during communication. Section II has the overview of related existing works. In
Cryptography has a long history. Its evolution and various Section III, proposed technique is presented. Experimental
methods are available in [1]. With advancement in results are presented in Section IV while Section V concludes
technologies, it is not enough if the contents of information the paper with future scope.
are kept secret as various Cryptanalysis techniques [2], [3]
have been developed to detect the secret information. Thus,
II. RELATED WORK
keeping the existence of the information secret is necessary.
This is achieved using Steganography. Steganographic embedding is not a new technique and has
been used since very long. Many researches have carried
Steganography is the art and science of hiding messages various works in this field till date. We present an overview
in such a way that no one except the sender and intended of few works carried out by others in this Section.
recipient suspects the existence of the message. The existence
of information is kept secret in Steganography by embedding Requirements of Steganographic system and survey of
the information in cover object. A basic block diagram of a various existing Steganographic systems are in [4], [8], [10].
Steganographic system is shown in Fig 1. The file used as Various existing techniques are analyzed in [11]. The
cover object varies depending on type of media used for transform domain Steganographic techniques such as DCT
Steganography [4]. In our method, image is chosen as cover. and DWT are discussed and analyzed in [12]. The security of
few Steganographic approaches are compared and analyzed

Volume 4, Issue 22 Published by, www.ijert.org 1

Special Issue - 2016 International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
ICACT - 2016 Conference Proceedings

in Shaveta Mahajan et al. [13]. The major topic of all these B. Discrete Cosine Transform
papers are various techniques that embed the data.
Discrete Cosine Transform (DCT) [25] is used to
Least Significant Bit is one of the simple Spatial Domain transform the image from spatial domain to frequency
techniques and is commonly used technique to embed the domain. The cover image of size × is first divided into
data. [14] uses LSB technique to embed the data. The data 8×8 blocks and then DCT is applied on each block. DCT-II is
bits are embedded only in selected areas obtained using pixel used for two dimensional images of M×N. The equation for
selection filter. Immunity to noise and compression is poor DCT-II is given by:
using this method. Shailender Gupta et al. [15] uses one, two,
three and four LSBs to embed the data once it is encrypted.
But data embedded using this technique can be easily
decoded. LSB techniques are simple and easy to detect. The
robustness of these methods are low and hence other methods eq (2)
were explored. where
Transform domain techniques are more immune to noise
and variations. [17] proposes a DCT based technique where
bits are embedded in DCT coefficients which are below
certain threshold. Takeshi et al. [18] proposes a DCT based and
algorithm that embeds data with less distortion. A Shield
algorithm based on DCT to embed the data is proposed in
[19]. DCT, a transform domain technique is robust but the
for x=1 to 8 and y=1 to 8
amount of information that can be embedded using this
technique is less.
On applying DCT on a block of image matrix I, a matrix
An embedding technique that uses both LSB and DCT is K is obtained. Matrix I has pixels as its entries whereas
proposed in [20]. Analysis and results of LSB and DCT matrix K has DCT coefficients as its entry. The obtained
method is discussed in [21]. These methods use advantages of matrix K has higher values at the top left. This higher value
both LSB and DCT technique. indicates that it corresponds to lower frequencies and higher
signal energies. The values other than that are very small
In [22], a method to embed multiple images with bit plane
enough such that it can be neglected with no or less
slicing is proposed. The imperceptibility of this algorithm
distortion.
goes on decreasing as the payload capacity is increased.
DWT based method to hide multiple images is proposed in C. Quantization
[23]. This approach is susceptible to noise. In general, DCT
method is more robust and imperceptible compared to other Quantization [28] is done to compress the DCT
methods proposed. coefficients obtained in previous step. Since human eyes are
insensitive to high frequency components, these are made
zero using standard Quantization matrix [24] which is
III. PROPOSED METHOD
The main aim of this work is to maximize the payload
capacity without compromising imperceptibility. Proposed
method targets grayscale images. Up to eight images can be
embedded in a single image using the proposed technique
which is discussed in this section. Selection of images (III-A)
is discussed first, sub-sections III-B to III-F discusses the
proposed LSB-DCT technique for embedding and finally
extraction of hidden data is presented in III-G.
A. Image selection
When only one LSB of each DCT coefficients of cover
image of size M×N is used to embed the data, the amount of Quantization is done by dividing each values in matrix K
information that can be embedded in the cover image is M*N with corresponding values in quantization matrix Q. This will
bits. Considering each pixel is represented using eight bits, result in matrix P, which has the quantized DCT coefficients.
the ideal size of a single image that can be embedded is X×Y, P(i, j) = K(i, j) / Q(i, j) eq (3)
where

1*X*Y*8=M*N eq (1) D. Zigzag scanning

The matrix P, obtained in previous step may have intra-
To embed a single image, all eight bits of pixels in the block correlation [9]. To remove this, zigzag scanning is done
secret image can be used. Embedding more than one secret on matrix P which results in one-dimensional array A. The
image uses only the MSBs of pixels in secret image. The values in this array will be in ascending order which means
number of MSBs used depends on the number of images to that low frequency elements are in the beginning.
be hidden. Four, two and one MSB(s) are selected to embed
two, four and eight images respectively.

Volume 4, Issue 22 Published by, www.ijert.org 2

Special Issue - 2016 International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
ICACT - 2016 Conference Proceedings

E. Embedding C. Mean Squared Error

Secret image is embedded by altering the LSB of the Mean Squared Error (MSE) [26] gives the average of the
values in the array A. Value in A is made even if the data bit squares of errors between cover image and stego image. It is
is even, else odd. calculated by
For embedding multiple images, LSBs of the data are
neglected since it carries less information. To embed two
images, four MSBs are considered and other four bits are
neglected. Thus it satisfies eq (1) with 2*X*Y*4=M*N where
2 is the number of images and 4 is the number of bits. Four eq (4)
images can be embedded by considering only two MSBs, where I is the cover image and S is the stego image.
neglecting the remaining six bits. Thus eq (1) becomes MSE for various images with different payload capacity
4*X*Y*2=M*N. Similarly 8 images can be embedded by for embedded and extracted images are listed in Table 1 and
considering only one MSB. Single image is embedded by Table 2.
using all bits in its pixels. D. Peak Signal to Noise Ratio
F. Image construction Peak Signal to Noise Ratio (PSNR) [27] is a measure of
After embedding, inverse zigzag is applied on array A to peak error. It is the ratio between the maximum possible
convert it back to the 8×8 matrix. Then this block is power of a signal to the power of corrupting noise that affects
dequantized by multiplying each value in matrix with the the precision, in decibels (dB). PSNR measures the quality
corresponding value in quantization matrix Q and Inverse between two images. It is given by
Discrete Cosine Transform (IDCT) is applied to transform it
back to spatial domain. Finally all 8×8 blocks are combined
to form the stego image S. eq (5)

G. Extraction where MAX is the maximum pixel value of image which is

255 for grayscale image.
Extraction of secret images from stego image requires the
stego image to be broken into 8×8 blocks, transform it to PSNR values for various images with different payload
frequency domain by applying DCT, quantize the capacity for embedded and extracted images are recorded in
coefficients, perform zigzag scanning and extract the data. Table 1 and Table 2.
The extracted bit is 0 if the value is even, else the extracted
bit is 1. Concatenate 8 bits together to form a pixel. Table 1. PSNR values of stego image with different capacity
for Fig 2.
When multiple images are hidden, the number of
embedded bits are extracted and is concatenated with 0 for Payload Capacity MSE PSNR (dB)
remaining bits to form the pixel. If two images are hidden, 1 0.5778 50.5127
four bits are extracted from stego image and is concatenated 2 0.5732 50.5476
with four zeros to form the pixel. These pixels are arranged to
4 0.5677 50.5894
reconstruct the secret image. Similarly four, eight images
hidden can be extracted. 8 0.5836 50.4695

IV. EXPERIMENTAL RESULTS Table 2. PSNR values of extracted image for different
There are many factors that decide how good a capacity of cover image when Fig 3a is hidden.
Steganographic system is. We consider few of these to Payload Capacity MSE PSNR (dB)
analyze our method. Results with each of the factors 1 1.0000e-03 78.1308
considered are discussed in this section. 2 77.4396 29.2412
4 1.3406e+03 16.8577
A. Imperceptibility 8 4.7548e+03 11.3595

Imperceptibility is the invisibility of the hidden image to

the human eye. This is the most important requirement of
Steganographic system since Steganography deals with
hiding data. The proposed method generates imperceptible
stego image in which hidden image cannot be noticed by
human eye. It is measured using PSNR which is discussed in
IV-D.
Fig 2. Cover Image
B. Payload capacity
Payload capacity [28] is the amount of information that
can be hidden in the cover image. The method we proposed
hides up to eight images in a cover image. Simulations are
carried out in MATLAB and obtained experimental results
are shown in Fig 2 to 7.
a b c d

Volume 4, Issue 22 Published by, www.ijert.org 3

Special Issue - 2016 International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
ICACT - 2016 Conference Proceedings

e f g h
Fig 3. (a)(b)(c)(d)(e)(f)(g)(h) Secret Images a

b c d e

a b

Fig 4. (a) Stego image with Fig 3a embedded

(b) Extracted image from Fig 4a f g h i
Fig 7. (a) Stego image with fig 3a to 3h hidden
(b)(c)(d)(e)(f)(g)(h)(i) Extracted images from Fig 7a

V. CONCLUSION AND FUTURE SCOPE

Steganography is a vast domain with its major application
a in watermarking and secret communication. It ensures that
the data is invisible and is protected from the third person.
There are various techniques of Steganography each with
own advantages and disadvantages. LSB method embeds data
directly in LSB of pixels of image whereas DCT method
embeds data in DCT coefficients of the image. Our proposed
method uses a mixture of both techniques to embed data
b c which gives good results with respect to factors analyzed.
Fig 5. (a) Stego image with fig 3a and 3b embedded
The main task was increasing the payload capacity of the
cover image. However, embedding more and more images
(b)and(c) Extracted images from Fig 5a decreases the quality of extracted images. When quality of
extracted image is not a concern, the proposed method can be
used.
Several further directions can be explored, including the
Steganalysis for the proposed approach. Further, payload
capacity can be increased if more LSBs are considered for
embedding. It would also be interesting to try embedding
images and texts together in a cover image using proposed
approach. Lastly, exploring the audio and video
a steganography with this technique is suggested.

REFERENCES
[1] Michael E. Whitman, Herbert J. Mattord, “Principles of Information
Security,” 4th Edition.
[2] Ashish Kumar Kendhe, Himani Agarwal, “A Survey Report on Various
Cryptanalysis Techniques,” IJSCE, vol 3, Issue 2, 2013.
b c d e [3] Vinod Saroha, Suman Mor, Jyoti Malik, “A Review of Various
Techniques of Cryptanalysis,” IJARCSSE, vol 2, Issue 10, 2012.
Fig 6. (a)Stego image with Fig 3a to 3d embedded
[4] Jasleen Kour, Deepankar Verma, “Steganography Techniques – A
(b)(c)(d)(e) Extracted images from Fig 6a Review Paper,” IJERMT, vol-3, Issue 5, 2014.
[5] Pooja Rai, Sandeep Gurung, M.K. Ghose, “Analysis of Image
Steganography Techniques: A Survey,” International Journal of
Computer Applications (0975-8887), vol 114, No. 1, 2015.

Volume 4, Issue 22 Published by, www.ijert.org 4

Special Issue - 2016 International Journal of Engineering Research & Technology (IJERT)
ISSN: 2278-0181
ICACT - 2016 Conference Proceedings

[6] S.G.Shelke, S.K.Jagtap, “Analysis of Spatial Domain Image [16] Ingemar J. Cox, Matthew L. Miller, Jeffrey A. Bloom, Jessica Fridrich,
Steganography Techniques,” IEEE International Conference on Ton Kalker, “Digital Watermarking and Steganography”, Morgan
Computing Communication Control and Automation (ICCUBEA), Kaufmann Publishers, 2nd Edition, 2008.
2015. [17] Hardik Patel, Preeti Dave, “Steganography Technique based on DCT
[7] Falesh M. Shelke, Ashwini A. Dongre, Pravin D. Soni, “Comparision Coefficients,” IJERA, vol. 2, Issue 1, 2012.
of different Techniques for Steganography in Images,” IJAIEM, vol 3, [18] Takeshi Ogihara, Daisuke Nakamura, Naokazu Yokoya, “Data
Issue 2, 2014. Embedding into Pictorial Images with Less Distortion Using Discrete
[8] Nan-I Wu, Chung-Ming Wang, Min-Shiang Hwang, “Data Hiding: Cosine Transform,” IEEE Proceedings of ICPR, 1996.
Current Status and Key Issues,” International Journal of Network [19] Deepika Bansal, Rita Chhikara, “An Improved DCT based
Security, vol 4, No. 1, pp. 1-9, January 2007. Steganography Technique,” IJCA, vol 102, No. 14, 2014.
[9] Chengjie Tu, Trac D. Tran, “Context-Based Entropy Coding of Block [20] Deepak Singla, Rupali Syal, “Data Security Using LSB & DCT
Transform Coeffiecients for Image Compression,” IEEE Transactions Steganography In Images”, IJCER, vol 2, Issue 2, 2012.
on Image Processing, vol 11, No. 11, 2002.
[21] Ekta Walia, Payal Jain, Navdeep, “An Analysis of LSB & DCT based
[10] R Poornima, R J Iswarya, “An Overview of Digital Image Steganography,” GJCST, vol 10, Issue 1, 2010.
Steganography,” IJCSES, vol 4, No. 1, 2013.
[22] H Faheem Ahmed, U Rizwan, “Embedding Multiple Images in an
[11] C.P.Sumathi, T.Santanam, G.Umamaheswari, “A Study of Various Image Using Bit Plane Slicing”, IJARCSSE, vol 3, Issue 1, 2013.
Steganographic Techniques Used for Information Hiding,” IJCSES, vol
4, No. 6, 2013. [23] Hemalatha S, U Dinesh Acharya, Renuka A, Priya R. Kamath, “A
secure and high capacity Image Steganography technique,” SIPIJ, vol
[12] Gurmeet Kaur, Aarti Kochhar, “Transform Domain Analysis of Image 4, No. 1, 2013.
Steganography,” IJSETT, 6(1): 29-37, 2013.
[24] Andrew B. Watson, “Image Compression Using the Discrete Cosine
[13] Shaveta Mahajan, Arpinder Singh, “A Review of Methods and Transform,” Mathematica Journal, 4(1), 1994.
Approach for Secure Stegnography,” IJARCSSE, vol 2, Issue 10, 2012.
[25] https://en.wikipedia.org/wiki/Discrete_cosine_transform.
[14] Rahul Joshi, Lokesh Gagnani, Salony Pandey, “Image Steganography
[26] https://en.wikipedia.org/wiki/Mean_squared_error.
With LSB,” IJARCET, vol 2, Issue 1, 2013.
[27] https://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio.
[15] Shailender Gupta, Ankur Goyal, Bharat Bhushan, “Information Hiding
Using Least Significant Bit Steganography and Cryptography,” [28] Jessica Fridrich, “Steganography in Digital Media”, Cambridge
IJMECS, 6, 27-34, 2012. University Press, 2010.

Volume 4, Issue 22 Published by, www.ijert.org 5