AN ENHANCED EDGE ADAPTIVE STEGANOGRAPHY APPROACH USING THRESHOLD VALUE FOR REGION SELECTION

International Journal on Computational Science & Applications (IJCSA) Vol.5,No.6, December 2015
DOI:10.5121/ijcsa.2015.5604 55
AN ENHANCED EDGE ADAPTIVE STEGANOGRAPHY
APPROACH USING THRESHOLD VALUE FOR REGION
SELECTION
Sachin Mungmode, R. R. Sedamkar and Niranjan Kulkarni
Department of Computer Engineering, Mumbai University, Mumbai, India
ABSTRACT
This paper attempts to improve the quality and the modification rate of a Stego Image. The input image
provided for estimating the quality of an image and the modified rate is a bitmap image. The threshold
value is used as a parameter for selecting the high frequency pixels from the Cover Image. The data
embedding process are performed on the pixels that are found with the help of Threshold value by using
LSBMR. The quality of an image is estimated by the value of PSNR and the modification rate of an image is
estimated by the value of MSE. The proposed approach achieves about 0.2 to 0.6 % of improvement in the
quality of an image and about 4 to 10 % of improvement in the modification rate of an image compared to
the edge detection techniques such as Sobel and Canny.
KEYWORDS
Steganography, Sobel Edge Detection, Canny Edge Detection, Mean Square Error (MSE), Peak Signal-to-
Noise Ratio (PSNR).
1.INTRODUCTION
Steganography has a Greece rootage where the word ‘Steganos’ means ‘Covered’ and ‘Graptos’
means ‘Writing’ [1]. Steganography is also known as ‘Secret Writing’. Steganography is a
subdivision of information security that hides the secret and crucial information into an innocent
medium in such a way that no one except recipient can extract the information [2]. The digital
medium can be an image, audio or video file [3]. In innovative life, steganography is working for
many purposes such as embedding copyright, watermark and covert communication. The
Steganalysis process targets to unwrap the presence of hidden messages in those stego medium.
The existence of steganalytic algorithms helps to find out the medium is a cover or not with a
more probability than random guessing.
Following valuable properties must be taken into circumstances while implementing
steganographic approaches.
1) Imperceptibility: Imperceptibility remains the firsthand destination of steganography. The
image must look exactly alike before and after the modification. A person must not be able to
see the changes in image.
2) Embedding Capacity: Capacity is the amount of information that can be embedded in a
particular medium. Capacity seems to be an issue for steganography as the size of the medium
and information to hide goes on changing. So, the embedding capacity remains a probatory
factor in steganography.

56
3) Robustness: Robustness is the level of difficulty required to demolish embedded information
without destructing the cover medium and extracting the required information.
4) Undetectability: Undetectability is the power of determining whether the cover medium
contains embedded information or not by employing different technical means.
Undetectability tries to hold back the fact that a message is being transferred from sender to
receiver.
Image steganography is the subdivision of steganography where digital images are used as bearer
file formats for information [4]. The Joint Photographic Experts Group Format (JPEG), Graphics
Interchange Format (GIF), Bitmap Picture format (BMP) and Portable Network Graphics format
(PNG) are the majority popular image file formats that are being shared on the internet.
Edge adaptive steganography is a peculiar case of the spatial or transform domain steganographic
technique [5]. The edge adaptive technique is based on area of interest where edge detection
techniques are used to choose a region randomly for data embedding. The edges of image are
detected using edge detection techniques like Sobel and Canny. As it selects the region of high
intensity in an image, the smooth areas stay unmodified. Advantage of this technique is the
quality of an image and the modification rate of an image improves.
This paper contains the results of execution of an enhanced technique on various images.
Following are the sections explained in the paper.
1) The first section of the paper gives the brief idea about steganography and related work on
Steganography.
2) The second section deals with the objective and the scope of the result, which are delivered by
execution of the algorithm.
3) The third section gives overview about Sobel edge detection technique, Canny edge detection
technique and LSBMR algorithm.
4) The fourth Section describes proposed technique for data embedding and data extraction.
5) The fifth section shows experimental results drawn by executing the proposed techniques on
nearly 100 images.
2.RELATED RESEARCH
The Least Significant Bit (LSB) is the basic technique used for steganography. LSB technique is
used to hide out the secret message bits into the least significant bits of the cover image. The LSB
is a swapping type of technique which swaps bits of cover image by secret message bit. Simple
LSB steganography is easily perceptible by steganalytic methods like Regular Singular (RS) and
Chi-square analysis [3].
LSBM (Least Significant Bit Matching) technique employed slight modification to the LSB
method [6]. In LSBM, if the private bit does not match the LSB of the cover image pixels, the
pixel values of an image are modified by adding one or subtracting one [7]. Statistically, the
probability of increasing or decreasing of pixel value for an image is the same and the imbalance
artifacts introduced in LSB substitution will be easily avoided. Therefore, the common
approaches that are utilized to determine the presence of LSB replacement are totally useless at
detecting the LSBM. The proposed method gives an option to either add or subtract one from the
cover image pixel at random. LSBM provided distortion and resistance to the Steganalysis [8].
The drawback of LSBM method is that the detection of the existence of the hidden messages
using the Centre of Mask component of the Histogram Characteristic Function [3] based detectors
is less efficient.

57
Jarno Mielikainen proposed LSBMR (Least Significant Bit Matching Revisited) which uses
grayscale cover images [9]. The data embedding is performed on the pair of pixels at a time.
LSBMR eliminated the asymmetry property caused by basic LSB. The Pixel-Value Differencing
(PVD) [10] technique uses different approach where the numbers of embedded bits are
determined by the difference between a pixel and its neighbor [11] [12]. Edge Adaptive LSBMR
(EALSBMR) proposed by Weiqi Luo, Fangjun Huang and Jiwu Huang.
Zohreh Fouroozesh and Jihad Al Jaam proposed Edge Adaptive Steganography using the Sobel
edge detection technique [13]. The technique is employed on grayscale images and LSBMR is
used for data embedding [1].
3.OBJECTIVE OF THE WORK
The Sobel and Canny edge detection techniques work well with gray scaled images. Therefore,
the color image need to be converted to a grayscale image for the data embedding process and
this image format conversion has a possibility of data loss. So, the improvised approach is carried
out on color images with LSBMR as a data embedding algorithm. The proposed mechanism aims
to improve the quality of an image and modification rate of stego image. PSNR and MSE are the
parameters for evaluating the quality of an image and the modification rate.
4.SOBEL EDGE DETECTION
The Sobel edge detector algorithm uses Sobel’s operator to extract edges [1]. For lower
embedding rates the sharp edges are used for holding information and as the rates increase the
algorithm adaptively is adjusted to use less sharp edges. Sobel operator is an orthogonal gradient
operator. It detects edges of the point, according to its adjacent points. The Sobel edge operator
performs a convolution in x-direction with the help of incline component called ‫ܩ‬௫ on selected
image. The Sobel edge operator performs a convolution in y-direction with the help of incline
component called G୷ on selected image. The incline components ‫ܩ‬௫ and G୷ are shown in figure 1.
+1 +2 +1
0 0 0
-1 -2 -1
‫ܩ‬௫ G୷
Figure 1. Incline components ‫ܩ‬௫ and G୷ with their masks
These kernels are designed to react maximally to edges running in vertical and horizontal
direction relative to the pixel grid, one kernel for each of the two perpendicular orientations [2].
The kernels are employed individually on input image to bring out individual measurements of
incline components ‫ܩ‬௫ and G୷. The combination of these two gives the absolute magnitude of the
incline at each point and the orientation of that incline. The incline magnitude is given by:
| ‫| ܩ‬ = ට‫ܩ‬௫
ଶ + ‫ܩ‬௬
ଶ (1)
Absolute magnitude of the incline is given by-
| ‫| ܩ‬ = |‫ܩ‬௫| + ห‫ܩ‬௬ห (2)
+1 0 -1
+2 0 -2
+1 0 -1

58
The angle of orientation of the edge giving rise to the spatial incline is given by-
ߠ = −tanିଵ
(‫ܩ‬௬/‫ܩ‬௫) (3)
Advantages:
• Provides a simple estimation to incline magnitude.
• Easy in detecting edges and their orientation.
Disadvantages:
• Sensitivity to the noise, edge detection and orientation of the edges.
• Inaccuracy to the incline magnitude.
5.CANNY EDGE DETECTION
The canny edge detector algorithm uses Gaussian operator for extracting edges of an image [6].
To lower embedding rate the sharp edges are used for holding information and as the rate
increases the algorithm adaptively is adjusted to use less sharp edges [7]. The three criteria used
by Canny for detecting edges are-
1) Low error rate- Every edge form the image should be detected and there should be no response
for non-edges.
2) Localization of edge pixels- The edge points of the image should be well localized. That
means, the distance between edge pixels as found by the detector and the actual edge is to be at
minimum.
3) One response to Single edge- There should be single response to single edge. This will
eliminate multiple responses to a single edge.
The Canny edge detection algorithm performs following steps in the edge detection process.
Step 1: Filtering of noise from the input image is carried out using Gaussian filter. Gaussian filter
calculates suitable mask and then Gaussian smoothing is done using standard convolution
methods.
Step 2: Find out the edge strength by taking the gradient of the image. The gradient of an input
image is estimated using a Sobel edge operator.
Sobel edge operator estimate gradient in x-direction with the help of the gradient component
called ‫ܩ‬௫ on selected image [1]. Sobel edge operator estimate gradient in y-direction with the help
of gradient component called G୷ on selected image. The gradient components ‫ܩ‬௫ and G୷ are
shown in figure 2.
-1 0 +1
-2 0 +2
-1 0 +1
‫ܩ‬௫ G୷
Figure 2. Gradient component masks in x-direction and y- direction
The absolute gradient magnitude or edge strength is calculated by using following formula-
| G | = |‫ܩ‬௫| + หG୷ห (4)
Step 3: The direction of the edge is computed using the incline in the x and y directions.
Whenever the value of ‫ܩ‬௫ is equal to zero, the edge direction has to be equal to 90
degrees or 0 degrees, depending on what the value of G୷. If G୷ has a value of zero, the
+1 +2 +1
0 0 0
-1 -2 -1

59
edge direction will be equal to 0 degrees. Otherwise the edge direction will be equal to 90
degrees.
The equation for finding the edge directions of an image is-
ߠ = −tanିଵ
(‫ܩ‬௬/‫ܩ‬௫) (5)
Step 4: Relate the edge direction to the direction of the image that can be traced. The four
directions that can be formed near to neighboring pixels are horizontal (0 Degrees), along
the plus diagonal (45 Degrees), perpendicular direction (90 Degrees) and along the minus
diagonal (135 Degrees).
Step 5: After knowing the edge directions, the edge thinning technique called non-maximum
suppression is applied. Non-maximum suppression is used to trace the edge along the
boundary of edges and conquer any pixel value that cannot be considered as part of an
edge. This will give thin line in an output image.
Step 6: Finally, hysteresis is being used to remove streaks. Streaking is the breaking up of an edge
outline that is caused due to fluctuation of operating above and below the threshold. For
hysteresis process requires two levels of threshold that are top level and bottom level
threshold.
Advantages:
• Finding of errors is effective and easy with the help of probability.
• Better Localization of edges and response for edges.
• Improving signal to noise ratio, as non-maximum suppression is used.
• Better edge detection in the noise state with the help of Thresholding method.
Disadvantages:
• Complex Computations.
• Time consuming.
6.LSBMR ALGORITHM
The Least Significant Bit Matching Revisited (LSBMR) [5] algorithm was used for gray-scaled
cover images. The algorithm uses two pixels of the cover image as embedding unit to hide out the
secret message. From the two pixels, the first pixel ‫ݔ‬௜ is used to conceal the secret message bit ݉௜
and the binary relationship between the pixels (‫ݔ‬௜, ‫ݔ‬௜ାଵ ) value is used to hide out another
message bit ݉௜ାଵ. The relationship between both pixels is calculated with the help of floor
function. The floor function is shown as follows:
݂(‫ݔ‬௜, ‫ݔ‬௜ାଵ) = ‫ݎ݋݋݈݂(ܤܵܮ‬ ቀ
௫೔
ଶ
ቁ + ‫ݔ‬௜ାଵ) (6)
LSBMR eliminates the asymmetry property caused by the basic LSB approach. It resists the
Steganalytic attacks that exploit the asymmetry property of the Stego-image. As it uses add one or
subtract one schema, the probability of the expected number of modifications reduced from 0.5 in
case of LSB to 0.375 for LSBMR for the same payload capacity. Hence, the statistical
detectability of the image steganography is low. LSBMR performs data embedding on pixel pairs
using following four cases:

60
Case 1: LSB (‫ݔ‬௜) = ݉௜& f (‫ݔ‬௜,‫ݔ‬௜ାଵ) = ݉௜ାଵ
(‫′ݔ‬௜,‫′ݔ‬௜ାଵ) = (‫ݔ‬௜,‫ݔ‬௜ାଵ) ;
Case 2: LSB (‫ݔ‬௜) = ݉௜& f (‫ݔ‬௜,‫ݔ‬௜ାଵ) ≠ ݉௜ାଵ
(‫′ݔ‬௜,‫′ݔ‬௜ାଵ) = (‫ݔ‬௜,‫ݔ‬௜ାଵ ± 1) ;
Case 3: LSB (‫ݔ‬௜) ≠ ݉௜& f (‫ݔ‬௜,‫ݔ‬௜ାଵ) = ݉௜ାଵ
(‫′ݔ‬௜,‫′ݔ‬௜ାଵ) = (‫ݔ‬௜ − 1,‫ݔ‬௜ାଵ) ;
Case 4: LSB (‫ݔ‬௜) ≠ ݉௜& f (‫ݔ‬௜,‫ݔ‬௜ାଵ) ≠ ݉௜ାଵ
(‫′ݔ‬௜,‫′ݔ‬௜ାଵ) = (‫ݔ‬௜ + 1,‫ݔ‬௜ାଵ) ;
Where ݉௜ and ݉௜ାଵ are message bits, ‫ݔ‬௜ and ‫ݔ‬௜ାଵ is a pixel pair before data embedding and
‫′ݔ‬௜ & ‫′ݔ‬௜ାଵ is a pixel pair after data embedding. The embedding cannot be performed for pure
pixels that have either a negligible or supreme allowable value. The message embedding rate is
discovered by the pseudo-random sequence generator procedure.
7.PROPOSED MECHANISM
The proposed scheme of edge adaptive steganography is based on the Threshold value and
LSBMR algorithm. The threshold value is the intensity difference between two consecutive
pixels; these two pixels are considered as an embedding unit for LSBMR. LSBMR takes message
bits and pixel pair as input for data encoding. The detailed data embedding algorithm are as
follows.
7.1.Data Embedding
Following figure 3 shows the block diagram of proposed technique used for data embedding.
Cover Image Shared Key
Secret Message
Stego Image
Figure 3. Data Embedding
The proposed data embedding technique are described in steps as follows:
Step 1: Separate RGB components of the cover image and choose the desired component (R, G,
and B) for data embedding.
Separate &
choose
RGB
component
Calculate
Threshold
as per
size of
message
Embedding
Region
Selection
Convert
message
to binary
bits
Data
embedding
using
LSBMR

61
Step 2: Convert the input message into binary bits with the help of ASCII value conversion.
Step 3: Evaluate the threshold value according to the size of secret message. The threshold value
is the intensity value difference between two pixels having difference greater or equal to
the threshold value.
Step 4: Select the region, according to the threshold value for generating embedding pairs. Save
the encoding pairs as secret key.
Step 5: The LSBMR algorithm is used for embedding the secret message bits into the cover
image.
Step 6: Rewrite the changes into cover image occurred during data embedding.
Step 7: Combine the RGB components of the cover image and save the image as Stego image.
7.2.Data Extraction
Following figure 4 shows the block diagram of proposed technique used for data extraction.
Stego Image Shared Key
Secret
Message
Figure 4. Data Extraction
The proposed data extraction technique is described in steps as follows:
Step 1: Separate the RGB components of Stego image and select the component from R, G, and B
that was used during data embedding process.
Step 2: Extract the intensities of encoding pairs using a shared key.
Step 3: Extract the message bits ݉௜ and ݉௜ାଵ from intensities that are calculated in step 2 and
using LSBMR algorithm.
Step 4: Convert binary bits of messages in ASCII value and then convert the ASCII value into its
character to get the original message.
8.EXPERIMENTAL RESULTS AND DISCUSSION
The tool used for implementing the work is MATLAB 7.6 and above versions. The objective of
the work is to extend the use of edge adaptive steganographic techniques for color images with
some modifications to the parameters used before and analyze the performance of the proposed
method. The proposed method is implemented on nearly 100 color images. The performance of
the proposed method is evaluated and compared on the basis of two parameters and the
Separate &
choose
RGB
component
Extract
intensities
of
embedded
pairs
Extract
the
message
bits
Convert message
bits into
characters

62
parameters are: Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR) and they are
computed as follows:
8.1.Mean Square Error
The mean square error (MSE) [1] is a statistical measure of how far modified values are different
from the actual values. It is mostly used in time series, but can be used in any sort of statistical
estimate [3]. Here in this project, it could be applied to pixel pairs, where one set is "Cover
Image" and the other is a “Stego Image”.
Steps for Calculating Mean Square Error:
1) Take Cover Image (CI) and Stego Image (SI) as inputs.
2) Subtract Stego Image values (pixel pair values) from Cover Image values.
3) Take the absolute value of each row. That is, if the difference is negative, remove the negative
sign. If it is positive, keep it as is.
4) Add up the absolute values of each row.
5) Take the square of the resultant.
6) Normalize the total number of rows and columns.ie. m x n.
MSE =
ଵ
୫ ×୬
∑ ∑ (CI − SI)ଶ୬ିଵ
୨ୀଵ
୫ିଵ
୧ୀଵ (7)
In above equation, CI= Cover Image, SI= Stego Image
In simple words MSE indicates average amount of modifications to the pixels.
8.2.Peak Signal-to-Noise Ratio
The peak signal-to-noise ratio (PSNR) [1] is the ratio between a signal's maximum power and the
power of the signal's noise. Here the PSNR is used to measure the quality of Stego image. Each
pixel of an image has a color value that may change after an image gets modified. Signals can
have a broad ever changing range; therefore PSNR is normally expressed in decibels [3]. Decibel
representation is in logarithmic scale. Peak Signal to Noise Ratio is computed using the formula:
ܴܲܵܰ = 10 × ݈‫݃݋‬ଵ଴ ቀ
ଶହ଺ ×ଶହ଺
ெௌா
ቁ (8)
The PSNR value indicates the quality of Stego image after modification of Cover image.The
MSE and PSNR parameters are calculated for several Cover images. Some of the Cover images
are taken as example for result calculation and comparison with existing techniques. Following
figure 5 considered as an example for the experiment.

63
(a) (b)
Figure 5. (a) Cover Image and (b) Stego Image
The results of MSE and PSNR for Cover Image figure 5 (a) and Stego Image figure 5 (b) using
different message lengths are shown in Table 1.
Table 1. MSE and PSNR values
Messag
e
Length
approxi
mately
(Bits)
MSE Values PSNR Values
Sobel Canny
Threshold
Value
Sobel Canny
Threshold
Value
400 0.000026 0.000026 0.000025 93.968707 93.968707 94.111111
600 0.000031 0.000031 0.000028 93.319299 93.319299 93.631876
900 0.000050 0.000050 0.000047 91.210765 91.210765 91.439394
1200 0.000060 0.000060 0.000054 90.400432 90.400432 90.820524
The table 1 shows four parameters for analyzing the results. First parameter is message length,
which is the length of data bits to be embedded inside the Cover image. Second parameter is an
edge detection technique for detecting edges of an image with the help of Sobel and Canny edge
detection methodology, and then LSBMR is implemented in order to embed the detected edges.
Third parameter is Mean Square Error for measuring the modification rate. The fourth parameter
is Peak Signal-to-Noise Ratio for measuring the quality of an image.Threshold Values and the
length of messages for figures 5 (a) and 5 (b) are shown in Table 2.
Table 2. Message length and Threshold value
Message length
approximately (Bits)
Threshold Value
400 170
600 164
900 151
1200 145
The table 1 shows improved values of PSNR and MSE parameters. From the results we can say
that the proposed Threshold Value mechanism is giving better results than Sobel and Canny edge

64
detection technique. The proposed approach achieves better quality of an image and reduces the
modification rate as compared to Sobel and Canny edge detection technique. The table 2 shows
the variation between threshold value and length of the message. The length of the message is
inversely proportional to a threshold value.
Another example is considered for evaluation of results to show the variations in the PSNR and
MSE values. Figure 6 consists of the cover image and stego image.
(a) (b)
Figure 6. (a) Cover Image and (b) Stego Image
The results of evaluation of MSE and PSNR values on Cover Image i.e. fig. 6 (a) and Stego
Image i.e. fig. 6 (b) using different message lengths is shown in Table 3.
Table 3.MSE and PSNR values
Message
Length
approxi.
(Bits)
MSE Values PSNR Values
Sobel Canny
Threshold
Value
Sobel Canny
Threshold
Value
400
0.00002
8
0.00002
8
0.000024
93.63187
6
93.63187
6
94.333875
600
0.00003
0
0.00003
0
0.000025 93.44164
3
93.44164
3
94.184104
900
0.00004
8
0.00004
8
0.000047
91.36183
9
91.36183
9
91.380765
1200
0.00006
4
0.00006
0
0.000059
90.10285
4
90.10285
4
90.431343
Threshold Values and the length of messages for figures 6 (a) and 6 (b) are shown in Table 4.
Table 4. Message length and Threshold value
Message length
approximately (Bits)
Threshold Value
400 119
600 113
900 101
1200 95

65
The table 3 shows improved values of PSNR and MSE parameters. From the results we can say
that the proposed Threshold Value technique is giving better results than Sobel and Canny edge
detection technique. The proposed technique provides better quality of an image as well as the
modification rate of an image gets reduced as compared to Sobel and Canny edge detection
technique. The table 4 shows the variation in Threshold Value in comparison with the length of a
message. The length of the message is inversely proportional to a threshold value. That means, as
the length of the message increases the threshold value gets decreases.
9.CONCLUSION AND FUTURE SCOPE
First the quality of an image and modification rate for an image is estimated using Sobel and
Canny edge detection techniques. Then the quality of an image and modification rate for an image
is estimated using proposed Threshold Value approach. The quality of an image is measured by
parameter PSNR and Modification rate for an image is measured by parameter MSE for all the
mentioned methods. The conclusions are carried out by considering the image in figure 5.
Figure 7. PSNR values of an Image
Figure 8. Percentage improvement in the quality of an image
The proposed Threshold Value approach results achieve about 0.2 to 0.6 % of improvement in the
quality of an image as compared to Sobel and Canny based technique. The percentage
improvement results in the quality of an image are shown in figure 8.
88
89
90
91
92
93
94
95
400 600 900 1200
Sobel
Canny
Threshold
Message Length
PSNRValues
0
0.1
0.2
0.3
0.4
0.5
400 600 900 1200
% Improvement
Message Length
Improvement(%)

66
Figure 9. MSE values of an Image
Figure 10. Percentage improvement in the modification rate of an image
Similarly, the proposed Threshold Value approach achieves 4 to 10 % of improvement in the
modification rate of an image as compared to Sobel and Canny based approach. The percentage
improvement results in the modification rate of an image are shown in figure 10.
Figure 11. Threshold Value Vs Message Length
Threshold Value is inversely proportional to the length of the message. That means, as the length
of the message increases the threshold value decreases. It helps in embedding number of data bits
during data embedding process. Figure 11 shows results variation between Threshold Value and
Message length.
0
0.00001
0.00002
0.00003
0.00004
0.00005
0.00006
0.00007
400 600 900 1200
Sobel
Canny
Threshold
Message Length
ModificationRate
0
2
4
6
8
10
12
400 600 900 1200
% Improvement
Message Length
Improvement(%)
130
140
150
160
170
180
400 600 900 1200
Threshold
Message Length
Threshold

67
The proposed Threshold Value approach can be implemented in the future for following real
world applications:
1) The proposed approach can be implemented to other cover medium such as audio/video.
2) The proposed approach can be applied on gray scale images.
3) The proposed approach can be used for frequency domain, type of data embedding.
4) The proposed approach can be implemented for real time Image Authentication System.
REFERENCES
[1] Zohreh Fouroozesh and Jihad Al jaam, (2014) “Image Steganography based on LSBMR using Sobel
Edge Detection”, IEEE.
[2] B. Sharmila and R. Shanthakumari, (2012) “Efficient Adaptive Steganography for color images
based on LSBMR algorithm”, ICTACT Journal on image and video processing, volume: 02, issue:
03.
[3] Weiqi Luo, Fangjun Huang and Jiwu Huang, (2010) “Edge Adaptive Image Steganography Based on
LSB Matching Revisited”, IEEE Transactions on Information Forensics and Security, vol. 5, no. 2.
[4] Nidhi Grover and A. K. Mohapatra, (2013) “Digital Image Authentication Model Based on Edge
Adaptive Steganography”, Second International Conference on Advanced Computing, Networking
and Security.
[5] J. Mielikainen, Jarno. (2006) "LSB matching revisited", Signal Processing Letters, IEEE 13, no. 5,
285-287.
[6] Bill Green, Canny Edge Detection Tutorial, 2002.
[7] Mamta Juneja, Parvinder Singh Sandhu, (2009) “Performance Evaluation of Edge Detection
Techniques for Images in Spatial Domain”, International Journal of Computer Theory and
Engineering, Vol. 1, No. 5.
[8] Syed Sameer Rashid, Swati R. Dixit and A. Y. Deshmukh, (2014) “VHDL Based Canny Edge
Detection Algorithm”, International Journal of Current Engineering and Technology, Vol.4, No.2.
[9] Nitin Jain, Sachin Meshram, Shikha Dubey, (2012) “Image Steganography Using LSB and Edge
Detection Technique”, International Journal of Soft Computing and Engineering (IJSCE) ISSN:
2231-2307, Volume-2, Issue-3.
[10] Yang C. H, Weng C. Y, Wang S. J, and Sun H. M , (2008) “Adaptive data hiding in edge areas of
images with spatial LSB domain systems”, Transaction on Information Forensics Security, Vol. 3,
No. 3, pp. 488–497.
[11] Fridrich J., Goljan J., and Du R., (2001) “Detecting LSB steganography in color, and gray-scale
images”, IEEE Multimedia, Vol. 8, No. 4, pp. 22–28.
[12] Deepali Singla and Mamta Juneja, (2014) “An Analysis of Edge Based Image Steganography
Techniques in Spatial Domain”, Proceedings of 2014 RAECS UIET Punjab University Chandigarh,
06 – 08.
[13] Zhenhao Zhu, Tao Zhang, and Baoji Wan, (2013) “A Special Detector for the Edge Adaptive Image
Steganography Based on LSB Matching Revisited”, 10th IEEE International Conference on Control
and Automation (ICCA) Hangzhou, China, June 12-14.

AN ENHANCED EDGE ADAPTIVE STEGANOGRAPHY APPROACH USING THRESHOLD VALUE FOR REGION SELECTION

More Related Content

What's hot (20)

Similar to AN ENHANCED EDGE ADAPTIVE STEGANOGRAPHY APPROACH USING THRESHOLD VALUE FOR REGION SELECTION (20)

Recently uploaded (20)

AN ENHANCED EDGE ADAPTIVE STEGANOGRAPHY APPROACH USING THRESHOLD VALUE FOR REGION SELECTION