2016 IEEE International Conference on Advances in Computer Applications (ICACA)

FPGA Implementation of Image Steganography using

Haar DWT and Modified LSB Techniques
Nikhil Simha H.N. Pradeep M. Prakash
Dept of Telecommunication Engineering, Dept of Telecommunication Engineering
Institute of Technology Bangalore, India Institute of Technology
[email protected] Bangalore,India
[email protected]
Suraj S. Kashyap
Dept of Telecommunication Engineering, Bangalore Sayantam Sarkar
Institute of Technology Dept of Electronics and Communication Engineering
Bangalore, India Vijaya Vittala Institute of Technology
[email protected] Bangalore, India.
[email protected]

Abstract—Security places an important role in Steganography is the technique in which we can embed a
communication applications for secure data transfers. Image secret image into a cover image without affecting its
Steganography is one of the most reliable technique in encryption perceptual quality of the cover image in such way that secret
and decryption of an image (hidden) inside other image (cover) image can be revealed by some process. The advantage of
such way that only cover image is visible. In this paper frequency Steganography over Cryptography is the inseparability of the
domain Image Steganography using DWT and Modified LSB stegano (secret image) from the cover image. Steganography
technique is proposed. The proposed approach uses DWT to schemes can be applied to video, audio and image based on
convert spatial domain information to frequency domain the application [3].
information. The LL band is used for further Image
Steganographic process. The image is decoded using inverse LSB. Steganography schemes are mainly divided into
Since the LL band is used for encoding and decoding purpose, following methods [4] as spatial-domain Steganography and
memory requirement of the design is less for hardware frequency-domain steganography. In spatial domain
implementation. Also this will increase the operating frequency steganography method the secret information is modified by
of the architecture. The proposed technique obtains high PSNR using spatial characteristics and then the modified information
for both stegano and recovered hidden image. is attached with carrier information to maintain the secrecy.
Normally pixel values and statistical trails are used to modify
Keywords— Haar DWT; Image Steganography; Modified LSB
the spatial cheracterstics of the secret message. Similarly in the
Technique; Gaussian Filter
frequency domain steganography method both the secret
I. INTRODUCTION message and carrier data are converted into frequency domain
by using domain transforms such as Discrete Fourier
Security is an important issue in various field such as Transform (DFT), Discrete Cosine Transform (DCT), and
communication and storage of confidential data in present era. Discrete Wavelet Transform (DWT), Fourier Mellin Transform
A variety of encryption scheme have been proposed to ensure (FMT), Fractal Transform etc. Then those frequency domain
security of data like cryptography, steganography etc. Image information is merged to generate stegano media. In decryption
encryption has wide applications [1] such as internet time domain transformation is necessary to extract the hidden
communication, multimedia systems, medical imaging, data.
telemedicine, military communication, etc.
Contribution- In this paper novel LSB technique is
Cryptography is used to achieve confidentiality (amongst proposed. The proposed technique takes 4 bits of binary hidden
others) in transmission and storage of data. Complex image pixel data from MSB side and merges with 4 bit MSBs
mathematical operation is performed by cryptographic of Cover image. This will increase the capacity of hidden
schemes for encoding and decoding secret data. The Visual image size. Also the uses of pre-shared key as the position of
Cryptography is mainly based on the vision systems of human the pixel of secret image will provide security.
being. This approach can be able to solve various security
issues with less complex mathematical models. In Visual Organization- In this paper Section-2 gives brief review of
Cryptography the secret information is hidden inside a visual existing techniques with their advantages and drawbacks.
media such as image or video which acts as carrier of the Section-3 presents the proposed architecture along with all
message. Visual cryptography scheme [2] is a secret sharing details. Simulation results and performance analysis are given
technique used for encrypting images. In this case the image is in Section-4. Similarly Section-5 and Section-6 represents
spitted into a large number of shares and at the decryption a FPGA implementation results and Performance comparisons
finite amount of share required. respectively.

978-1-5090-3770-4/16/$31.00©2016 IEEE
26

Authorized licensed use limited to: R V College of Engineering. Downloaded on April 19,2025 at 04:52:58 UTC from IEEE Xplore. Restrictions apply.
 2016 IEEE International Conference on Advances in Computer Applications (ICACA)

II. LITERATURE SURVEYS DWT is applied and only LL band is considered. This is
Ran-Zan and Yeh-Shun [5] proposed two way block because the LL band contains most of the valuable
matching based Steganography technique. This technique first information about the image and the size of LL band is one
generates a series of blocks and then search the similar blocks forth of actual image which helps to reduce the size of the
from the image. Hop embedding schemes are used to merge memory and channel bandwidth. As a result we get faster
secret information into cover image. This scheme provides processing. By taking both LL bands the stegano image is
high PSNR for both cover and Hidden information. Vojtech generated by using modified LSB techniques. This stegano
and Jessica [6] proposed high pass directional filter banks with image is transmitted through communication channel and at
adaptive steganographics distortion function. The distortion is the receiving end reverse operation is performed to retrieve
used to obtain the directional residues which are mainly used to back LL band of hidden image.
measure the quality of stegano image. Chen et al., [7] proposed
adaptive embedding algorithm for image Steganography. To
minimize error multi-image and multi-bits are used by the
algorithm. The algorithm is mainly divided into three steps as
(i) embedded logo into cover (ii) adjust LSB of cover
adaptively (iii) adjust MSB of cover adaptively. Ajit and
Manjula [8] proposed an image steganography using DWT and
hybrid wavelet transform. In this method first both images are
normalized and thed DWT is applied on the respective image
to generate all four sub-bands. Next all sub-bands are fused to
get stegano image. Manoj kumar et al., [9] proposed image
steganography which is mainly based on the Data Encryption
Standard (DES) algorithm. The DES algorithm uses the S-box
mapping and secret key. Embedding function is used to pre-
process the secret image. Now by replacing the embedding
function values into the cover image the stegano image is
formed. Prabhakaran and Bhavani [10] proposed image
steganography technique using Discrete Wavelet Transform
(DWT). Amold transformation with secret key is used to make
secret image. Then DWT is applied to both cover image and
payload followed by alpha blending operation. This alpha
blending matrix is obtained by addition of wavelet sub bands of
cover image and secret image. Payload is hidden in the DWT
coefficient of cover image. Prem kumar and Narayanan [11]
have proposed a new scheme for secure banking application
based on visual cryptography. The proposed technique
incorporates both steganography and cryptography technology.
Maximum numbers of surrounding pixels are considered to
achieve best embedding capacity of every target pixel. Rong–
Jian and Shi-Jinn [12] proposed an anti-forensic steganography
system having high performance and high embedding capacity.
To increase the randomness of encoding and decoding multi-bit
adaptive embedding concept is used. Chao wang et al., [13]
proposed fast matrix embedding algorithm for image
Steganography. This method based on matrix extending for
reducing the computational complexity of matrix based
embedding. Vladimar et al., [14] proposed a secure
steganography system in JPEG file based on modulus function
which is secure against histogram attacks. First the message is
encrypted using 128 bit AES cipher which will give added
security. This encrypted text is then embedded in a JPEG
image.
III. PROPOSED ARCHITECTURE
The basic block diagram of proposed architecture is
divided into two parts encoding and decoding as shown in the
Fig. 1 and Fig. 2 respectively. At encoding part first both the
image is pre-processed and then noise present in the image is
filtered out by Gaussian filter. This will increase the quality of
stegano and recovered image. On the filtered image Haar Fig. 1. Encoding of Two Images

27
Authorized licensed use limited to: R V College of Engineering. Downloaded on April 19,2025 at 04:52:58 UTC from IEEE Xplore. Restrictions apply.
 2016 IEEE International Conference on Advances in Computer Applications (ICACA)

equations are implemented on hardware by using adder,

subtractior and shifter only.

= (3)
= (4)
= (5)
= (6)

Where, is 2x2 image sub-matrix.

yLL, yLH, yHL and yHH are four sub-bands respectively.

Fig. 2. Decoding of Hidden Image

A. Pre-processing The Haar wavelet decomposition of Lena image is shown in

In pre-processing first the image is converted to a specific Fig. 3 below.
size (i.e. 256x256) which is suitable for hardware
implementation. Also color image is converted to gray image
to reduce hardware requirements.
B. Gaussian Filter
Gaussian filter is mainly used to remove random noise
present in any image. The equation for Gaussian filter in two
dimensions [15] is given in equation (1) below

( , )= (1)

Where, x is the distance from the origin in the horizontal

axis. y is the distance from the origin in the vertical axis. σ is
the standard deviation of the Gaussian distribution.
The 3x3 Gaussian filter equation [15] is then
Fig. 3. Haar Wavelet Decomposition
1 2 1
D. Modified LSB Technique
= 2 4 2 ∗ (2)
1 2 1 ℎ The modified LSB technique is divided into two parts
namely encoding and decoding as explained below
a) Encryption Algorithm
Where, is 3x3 sub-matrix pixel values of an In this case we hide one image inside other image using a
ℎ pre-shared key. So, the output is encrypted image which is
image. also known as stegano image. The pre-shared key is mainly
C. 2D-DWT used for security purpose which is shared by both encoding
and decoding end for proper operation. Here pre-shared key is
Discrete Wavelet Transform (DWT) gives multi-resolution used to generate the pixel position for encoding which will
characteristics of the image in frequency domain. The DWT is give more security.
used in the paper is Haar DWT.
Consider the last 4 bits from MSB side of both cover and
We can decompose an image having MxN number of hidden image. Now merge both image pixels using
pixels into four sub-bands where each sub-bands having concatenate operation and the position of the pixel is decided
(M/2)x(N/2) pixels. The four sub-bands are named as LL, LH, by the pre-shared key.
HL and HH bands [16] respectively. Most of the information
(low frequency components) of the image is present in LL The steps involved in encryption is given below
bands and remaining bands contains edges (high frequency i. Take the pixel values of cover image in binary format
components) of the images. present in the image.
ii. Consider only last 4 bits of the corresponding pixel
The equation for first four sub-bands [16] of Haar
transform is given in equation (3) to equation (6) below. Those value in MSB side.

28
Authorized licensed use limited to: R V College of Engineering. Downloaded on April 19,2025 at 04:52:58 UTC from IEEE Xplore. Restrictions apply.
 2016 IEEE International Conference on Advances in Computer Applications (ICACA)

iii. Take the pixel value of the hidden image whose

position is defined by the pre-shared key equation.
iv. Consider only last 4 bits of the corresponding pixel
value in MSB side.
v. Merge both truncated pixel values to generate
stegano image using the equation is given as

=
( ) ( ) (7)

b) Decryption Algorithm
This is reverse of encryption algorithm i.e. recover back
the hidden image from cover image. To decrypt the stegano
image first consider 4 bits from LSB side and then pad four
zeros after the number. Now store that number into a Fig. 4. Software Simulation Output
temporary memory. Using pre-shared key determine the pixel
values in actual order and send it to output which is recovering
of hidden image. C. Performance Analysis
The steps involved in decryption is given below
To analyze the performance of proposed architecture we
i. Take the pixel values of stegano image in binary
take different test images and encrypt same information into
format present in the image.
the image and check PSNR values using equation (7). Those
ii. Image pixel position for decryption is defined by pre-
values are tabulated into Table 1.
shared key equation.
iii. Consider only fast 4 bits of the corresponding pixel TABLE I. SIMULATED PSNR VALUE OF PROPOSED ARCHITECTURE
value in LSB side.
iv. Merge the truncated pixel values with 4 zeros to Cover Hidden PSNR Value (dB) PSNR value (dB) of
Image Image of Cover Image Hidden image
generate hidden image using the equation is given as
Lena Plane 75.25 71.83
=
( ) (4 ) (8) Barbara Rice 70.05 65.41

Mandrill RBC 73.75 70.38

Plane Lena 74.73 71.39

IV. SIMULATION RESULTS AND PERFORMANCE ANALYSIS Rice Barbara 70.54 70.02
In this section the proposed design is simulated using
RBC Mandrill 71.68 69.79
MATLAB R2012a (7.14.0.739) version.
A. Performance Parameter Taj Cameraman 69.68 65.62
Mahal
In this section, the performance parameters are evaluated
using PSNR values between original image and stegno image V. FPGA IMPLEMENTATIONS
by using the formula [17] as
In this section the proposed design is implemented on
Spartan 6 (XC6SLX45-3csg324) FPGA board and the coding
= 10 log (9) is done by using VHDL language.
A. Hardware Utilization
Where, MSE = ∑ ∑ I (x, y) − O (x, y) Device utilization of the architecture is given in Table 2.
IN(x,y) is input image pixel values. Gaussian Filter uses 112 slice registers, 225 slice LUTs, 29
OP(x,y) is stegno image pixel values. LUT-FF pairs and it’s maximum operating frequency is
AB is the image dimensions (256x256). 164.841 MHz. LSB encoding and decoding both uses
separately 3 slice registers, 1 slice LUTs and it’s maximum
B. Image Output operating frequency is 866.643 MHz. Similarly Haar DWT
The original, hidden, stegno and recovered image is shown uses 359 slice registers, 1835 slice LUTs, 248 LUT-FF pairs
in Fig.4 where the cover image is a Lena image and the hidden and maximum operating frequency is 286.75 MHz.
image is Barbara image. Total module uses 514 slice registers, 331 slice LUTs, 297
LUT-FF pairs and the maximum operating frequency is
153.312 MHz. The extra hardware resources are used to

29
Authorized licensed use limited to: R V College of Engineering. Downloaded on April 19,2025 at 04:52:58 UTC from IEEE Xplore. Restrictions apply.
 2016 IEEE International Conference on Advances in Computer Applications (ICACA)

maintain synchronous operation between all blocks. Also this most of the image information is present (ii) Single bit
reduces the operating frequency of overall module. replacement concept is replaced by multi-bit concept

TABLE II. HARDWARE UTILIZATIONS OF PROPOSED ARCHITECTURE TABLE III. PSNR COMPARISONS OF PROPOSED ARCHITECTURE WITH
EXISTING ARCHITECTURES
Parameters Gaussian LSB LSB Haar Total
Filter Encoding Decoding DWT Authors Techniques Cover PSNR
Image (dB)
No. of Slice 112 3 3 359 514 Motamedi and Wavelet transforms and Barbara 39.65
Registers Jafari [18] image de-noising Boat 36.34
No. of Slice 225 1 1 1835 2085 techniques.
LUTs Tasnuva LSB with Pixel value Mandrill 32.67
No. of fully 29 0 0 248 297 Mahajabin et difference.
used LUT- al.,[19]
FF pairs Ashish Soni et Discrete Fractional Rice 32.46
Maximum 164.841 866.643 866.643 286.759 153.3 al.,[20] Fourier Transform.
operating 12 Barbara 70.05
frequency Proposed Method Haar DWT and Modified Boat 72.49
(MHz) LSB Technique Mandrill 73.75
Rice 76.79
B. Image Output
The original, hidden, stegno and recovered image is shown B. Hardware Comparisons
in Fig. 5 where the cover image is a real JPEG image and the Hardware comparisons of proposed technique with
hidden image is cameraman image existing techniques are given in Table 4. The architecture
presented by Maya and Sabarinath [21] is implemented on
Spartan-3 EDK embedded processor and the architecture uses
1474 slice registers, 1569 slice flip-flops, 2330 LUTs and 3
MULT18x18s. The algorithm presented by Jatin and Bhatt
[22] is implemented on NIOS processor and the coding is
done using C++ language which uses 2411 slice registers and
maximum operating frequency is 107.75 MHz. The
architecture presented by Vasantha and Vidhya [23] is
implemented on Spartan-3 EDK embedded processor and the
architecture uses 1880 slice registers, 2118 slice flip-flops,
2971 LUTs and 3 MULT18x18s. The proposed Technique is
implemented on Spartan-6 FPGA and coding is done by using
VHDL language. The proposed architecture uses 514 slice
registers, 297 slice flip-flops, 2108 LUTs and maximum
operating frequency is 153.312 MHz. Hardware utilization is
very less compared to existing techniques because of simple
and multiplier-less architecture. Here multipliers are replaced
by shifters.
Fig. 5. Hardware Output TABLE IV. HARDWARE COMPARISONS OF PROPOSED ARCHITECTURE
WITH EXISTING ARCHITECTURES

Parameters Maya and Jatin Vasantha Proposed

VI. PERFORMANCE COMPARISONS Sabarinath and and Videya Method
[21] Bhatt [23]
In this section performance comparison of proposed [22]
technique with existing techniques in-terms of both software Device Spartan-3 Nios-II Spartan-3 Spartan-6
EDK Processor EDK
(i.e. PSNR) and hardware (i.e. device utilization and Processor Processor
maximum operating frequency) is discussed
No. of Slice 1475 2411 1880 514
A. Software Comparisons Registers
Table 3 shows the comparison of PSNR of proposed No. of Slice 1569 ---- 2118 297
Flip-flops
technique and the existing techniques. The PSNR varies
between 35.79 and 39.65 based on cover images. The No. of 4 input 2330 ---- 2971 2108
LUTs
proposed technique is compared with existing techniques
presented by Hoda Motamedi and Ayyoob Jafari [18], No. of 3 ---- 3 0
MULT18x18s
Tasnuva Mahajabin et. al., [19] and Ashish Soni et al., [20]. It Maximum ---- 107.75 65.342 153.312
is observed that the PSNR values values are higher in the case operating
of proposed algorithm compare to existing algorithms for the frequency
following reasons as (i) only LL band is considered where (MHz)

30
Authorized licensed use limited to: R V College of Engineering. Downloaded on April 19,2025 at 04:52:58 UTC from IEEE Xplore. Restrictions apply.
 2016 IEEE International Conference on Advances in Computer Applications (ICACA)

VII. CONCLUSION [9] Manoj Kumar Ramaiya, Naveen Hemrajani and Anil Kishore Saxena,
“Security Improvisation in Image Steganography using DES,” 3rd
Image steganography is the current area of research. In International Advance Computing Conference, pp. 1094 – 1099,
existing techniques both DWT and IDWT is used to provide February 2013, India.
security at decryption stage. But this will need large amounts [10] Prabakaran G and Bhavani R, “A Modified Secure Digital Image
of hardware resources and large execution time. To reduce Steganography Based on Discrete Wavelet Transform,” International
Conference on Computing Electronics and Electrical Technologies, pp
hardware utilization and execution time the proposed system .1096 - 1100, March 2012, India.
uses only LL band of both cover and hidden image for
[11] S Prem Kumar and A E Narayanan, “New Visual Steganography
encryption. The uses of pre-shared key provide high security Scheme for Secure Banking Application,” IEEE International
for both encryption and decryption. Also in proposed Conference on Computing Electronics and Electrical Technologies,
architecture multiplier and dividers are replaced by shifters pp.1013-1016, March 2012, India.
which will increase the operating frequency. The proposed [12] Rong - Jian Chen and Shi-Jinn Horng, “Multi-Bit Adaptive Embedding
approach involves frequency domain transformation for Algorithm for Anti Forensic Steganography,” IEEE International
compression of image which combined with modified LSB Symposium on Biometrics and Security Technologies, pp.82-89, March
2012, Taiwan.
scheme provides better PSNR for any image.
[13] Chao Wang, Welming Zhang, Jiufen Liu and Nenghai Yu, “Fast Matrix
Embedding By Matrix Extending,” IEEE Transactions on Information
Forensics and Security, Vol. 7, No 1, pp. 346-350, February 2012.
[14] Vladimir Banoci, Gabriel Bugar, Dusan Levicky and Zita Klenovicova,
References “Histogram Secure Steganography System in JPEG File Based on
[1] Sagar Kumar Nerella, Kamalendra Verma Gadi and Raja Sekhar Modulus Function,” Twenty Second International conference
Chaganti, “Securing Images Using Color Visual Cryptography and Radioelectronika, pp. 1 – 4, April 2012, Czech Republic.
Wavelets”, International Journal of Advanced Research in Computer [15] [Online]https://en.wikipedia.org/wiki/Gaussian_filter.
Science and Software Engineering, Vol. 2, Issue. 3, pp. 163-168, March [16] Satish S Bhairannawar, Anand R, Raja K B and Venugopal K R, “FPGA
2012. Implementation of Fingerprint Recognition System uing Adaptive
[2] Ram Krishna Jha and Abhijit Mustafi, “Boolean XOR Based (K, N) Threshold Technique”, IEEE International Conference on Electrical,
Threshold Visual Cryptography for Grayscale Images”, International Electronics, Signals, Communication and Optimization, pp. 1-5, January
Journal of Computer Science and Informatics, Vol. 2, Issue. 3, pp. 42- 2015, India.
45, 2012. [17] [Online]https://en.wikipedia.org/wiki/Peak_signal-to-noise_ratio.
[3] Jagdeep Verma and Vineeta Khemchandani, “A Visual Cryptographic [18] Hoda Motamedi and Ayyoob Jafari, “A New Image Steganography
Technique to Secure Image Shares”, International journal of based on Denoising Methods in Wavelet Domain,” 9th International
Engineering Research and Applications, Vol. 2, Issue. 1, pp.1121-1125, Conference on Information Security and Cryptology, pp. 18 – 25,
2012. September 2012, Iran.
[4] B. Pushpa Devi, Kh. Manglem Sing and Sudipta Roy, “Dual Image [19] Tasnuva Mahjabin, Syed Monowar Hossain, and Md. Shariful Haque,
Watermarking Scheme based on Singular Value Decomposition and “A Block Based Data Hiding Method in Images using Pixel Value
Visual Cryptography in Discrete Wavelet Transform”, International Differencing and LSB Substitution Method,” 15th International
Journal of Computer Applications, Vol. 50 , No.12, pp. 7-12, July 2012. Conference on Computers and Information Technology, pp. 168 – 172.
[5] Ran-Zan Wang and Yeh-Shun Chen, “High-payload Image December 2012, Bangladesh.
Steganography using Two-way block matching,” IEEE Signal [20] Ashish Soni, Jitendra jain and Rakesh Roshan, “Image Steganography
Processing Letters, Vol. 13, Issue 3, pp. 161 – 164, March 2006. using Discrete Fractional Fourier Transform,” International Conference
[6] Vojtech Holub and Jessica Fridrich, “Designing Steganographic on Intelligent Systems and Signal Processing, pp. 97 – 100, March 2013,
Distortion using Directional filters,” IEEE International Workshop on India.
Information Forensics and Security, pp. 234 – 239, December 2012, [21] Maya C S and Sabarinath G, “ An Optimized FPGA Implementation of
West Africa. LSB Replacement Steganography using DWT”, International Jurnal of
[7] Rong-Jian Chen, Jui-Lin Lai and Shi-Jinn Horng, “Novel Multi-bit and Advanced Reaserch in Electrical, Electronics and Instrumentation
Multi-image Steganography Using Adaptive Embedding Algorithms Engineering, Vol. 2, Special Issue. 1, pp. 586-593, December 2013.
with Minimum Error,” 5th International Conference on Innovative [22] Jatin Chaudhari and K. R. Bhatt, “FPGA Implementation of Image
Mobile and Internet Services in Ubiquitous Computing, pp. 221 – 228, Steganography:A Retrospective”, International Journal of Engineering
July 2011, South Korea. Development and Research, Vol. 2, Issue. 2, pp. 2117-2121, 2014.
[8] Ajit Danti and G R Manjula, “Secured Data Hiding of Invariant Sized [23] B. Vasantha Lakshmi and B. Videya Raju, “FPGA Implementation of
Secrete Image based on Discrete and Hybrid Wavelet transform,” IEEE Lifting DWT based LSB Steganography using Micro-Blaze Processor”,
International Conference on Computational Intelligence & Computing International Journal of Computer Trends and Technology, Vol. 6, No.
Research, pp. 1 – 6, December 2012, India. 1, pp. 6-14, December 2013.

31
Authorized licensed use limited to: R V College of Engineering. Downloaded on April 19,2025 at 04:52:58 UTC from IEEE Xplore. Restrictions apply.