Understanding Image/Video Steganography · allowing me to explore image/video steganography and steganalysis. Index Terms - Steganalysis, Steganography, Kerckhoff’s Principle, Prisoner’s

Abstract—Traditional steganography is the process of

selecting the appropriate digital media cover to conceal

secret information within this digital media. Steganalysis is

the process of detecting and understanding this

steganographic information. A popular formulation of the

steganography paradigm is the well-known “The

Prisoner’s Problem”. The intent of this article is to adjust

this traditional paradigm in two aspects. First, the

steganography does not have control over the cover image

selection that is used to embed the information. Second,

steganalysis objective is expanded to not only detect the

steganographic information but to effectively and

efficiently neutralize the steganographic information

within the cover image without significantly corrupting the

cover message. This paper explores steganalysis processes

that eliminates and/or disrupts the steganographic

information, while maintaining the quality of the cover

image. This paper explores spatial, frequency and time

domains. The author would like to thank Dr. Gaj for

allowing me to explore image/video steganography and

steganalysis.

Index Terms - Steganalysis, Steganography, Kerckhoff’s

Principle, Prisoner’s Problem, Embedding Messages,

Information Hiding, LSB (Least Significant Bit), Discrete

Cosine Transform (DCT), Encoding

I. INTRODUCTION

Steganography is a composite of the Greek words

“steganos”, meaning “covered” and “graphia” meaning

“writing”. Steganography is another term for covert

communications and is a technique for hiding information in

digital media. Whereas, steganalysis is the process of

detecting and understanding the embedded steganographic

message. Interest in steganography and steganalysis has been

increasing as evidence by the number of steganography

articles annually published by IEEE. This exponentially

increase in steganography interest mirrors the increasing

commercial communications bandwidths in support of larger

digital media demands.1

Steganographics Growth Data Communication Growth

Steganography has a triad relationship between embedding

capacity, undetectability and robustness. Capacity is the

maximum amount of secret information that can be

embedded into a cover file. Capacity is an absolute value in

terms of number of information bits that are embedded into

the cover image. Capacity value depends on both embedding

function and cover properties. For example, in the LSB

technique if the cover is an 8 bit grayscale image for one bit

per pixel embedding the capacity would be equal to 12.5%

bandwidth. Undetectability is defined as the steganographic

image should not have perceptual artifacts. This property

would be satisfied if difference of the resultant

steganographic image is not distinguishable from original

cover image. Robustness is a property of the difficulty of

eliminating secret information from the steganographic file.

Property of robustness talks about resisting against

intentional distortion of the communication channel by

means of systematic interface of channel noise aiming to ban

use of steganography technique.2,3

A popular construction of the steganography paradigm is the

well-known “The Prisoner’s Problem”.4 Where Alice and

Bob are imprisoned in separate cells and want to hatch an

escape plan. They are allowed to communicate but their

communication is monitored by Warden Eve. If Eve finds

out that the prisoners are secretly exchanging messages, she

will cut the communication channel and throw them into

solitary confinement. The prisoners resort to steganography

as a means to exchange the details of their escape. When Eve

discovers that Alice and Bob communicate secretly, the

steganographic system is considered broken. This is in

contrast to encryption, where a successful attack means that

the attacker gains access to the decrypted content or partially

recovers the encryption key. It is assume that Warden Eve

has a complete knowledge of the Steganographic algorithm

that Alice and Bob might use, with the exception of the secret

steganographic key, which supports Kerckhoff’s Principle

which states that security of the communication should not

lie in the secrecy of the system but only in the secret key.

This paper adjusts “The Prisoner’s Problem” to explore the

balance between neutralizing the steganographic message in

the cover image and maintaining the quality of the cover

image. “The ‘Modified’ Prisoner’s Problem” has Alice,

who is outside the prison and wants to send an escape plan

to Prisoner Bob. The only communication channel available

is the prison’s security video that watches the front gate. The

security video system is monitored by Warden Eve. Warden

Eve will arrest Alice if Warden Eve can ascertain that Alice

Understanding Image/Video Steganography Clair E. Guthrie

Graduate Student in Electrical and Computer Engineering at George Mason University ECE646 Cryptography and

Computer Network Security Project, IEEE

is secretly sending steganographic messages via the prison

security surveillance video to Bob. In addition, if Alice

shows up at the coordinated escape location and Bob does

not (i.e. the steganographic message was neutralized and

was never received by Bob) Warden Eve will arrest Alice.

However, Warden Eve requires quality video to monitor the

prison’s front gate. If she loses video quality the entire

prison population will escape (via the front gate). Warden

Eve may detect the presence of a steganographic message,

but has to neutralize any steganographic message that

security image/video may contain. It is assumed that

Warden Eve has complete knowledge of the steganography

algorithm, but not the secret steganographic key. Hence,

Warden Eve needs to balance the elimination of the

steganographic message (i.e. escape plan) against

maintaining prison situational awareness via video quality

(i.e. cannot turn-off or significantly degrade the prison

security video).

A straightforward example of Steganography and

Steganalysis is provided. Assume the Steganographic

Message is “HELLO” and needs to be embedded into a 4 x

6 grayscale image. First, “HELLO” is converted to ASCII

(72, 69, 76, 76, 79). Next sixty-five is subtracted from the

ASCII characters (72-65, 69-65, 76-65, 76-65, 79-65) which

yields (7, 4, 11, 11, 14). This is done to support encryption.

The steganographic algorithms determine encoding and

placement of the message into the Cover Image. In this

example, the Cover Image is a limited grayscale 4 x 6 pixels

(the numbers in the pixels represent the grayscale value (i.e.

0 = Black, 25 = White). The “green boxes” represent the

encoded message (e.g. steganographic pixel image located at

row 1 and column 3 is replaced with 7). The steganalysis

side (i.e. Warden Eve) has never seen the original cover

image. The “red boxes” represent the Steganalysis

neutralization that is trying to eliminate or disrupt the

message by randomly injecting random values into selected

pixels (e.g. steganographic image pixel located at row 1 and

column 6 is replaced with the number 25). The steganalysis

image is then delivered to the receiver (Bob), who pulls out

the encrypted message and then decrypts the message. The

appropriate steganographic pixel locations with their values

are extracted and sixty-five is added to this number,

providing an ASCII character. In this example the

neutralization is partially effective, transforming “HELLO”

into “HVLLA”. However, this steganalysis neutralization

approach has significantly affected the image quality by

changing 33% of the picture pixels (8 of the 24 pixels).5,6

II. DEFINITIONS

This paper defines the Cover Image as the prison

image/video used to conceal the message (escape plan) via

steganography.

The Steganographic image has embedded the encrypted

message (escape plan) in the cover image (prison

image/video).

The steganalysis image is the steganographic image that has

been altered/sanitized to try to eliminate or neutralize the

embedded steganographic message (i.e. escape plan).7

For the purpose of this paper and supported by Matlab

algorithms, the prison image/video is a 480 x 720 grayscale

pixel image. Video will be briefly addressed later in this

paper. Grayscale image’s pixel are shades of gray from 0

(Black) to 255 (White), with each pixel represented by 8

bits (i.e. one byte, 256 gray colors).8 True Color (Red,

Green Blue (RGB)) image’s pixel are described by the

amount of red, green and blue per pixel. Each of these

components (RGB) has a range 0-255, this gives a total of

16,777,216 different possible colors. The True Color image

is a “stack”” of 3 matrices, representing red, green and blue

values for each pixel (i.e. every pixel corresponds to 3

values). True Color was not used in this effort. However,

the results of this effort could be directly applied to True

Color using any or all three colors (RGB).

The image size used for this paper was Standard Definition

(SD) format. SD format is 480 by 720 pixels for a total of

345,600 pixels. The results of the efforts defined in this

paper could be applied to both High Definition images

(1920 by 1080 pixels for a total of 2,073,600 pixels) and 4K

Definition images (3840 by 2160 pixels for a total of

8,294,400 pixels) and their associated video rates (30/60

Hz).

row col

H 72 = 7 1 3 1 2 3 4 5 6

E 69 = 4 2 2 1 1 2 3 4 5 6

L 76 = 11 3 4 2 7 8 9 10 11 12

L 76 = 11 3 6 3 13 14 15 16 17 18

O 79 = 14 4 2 4 19 20 21 22 23 24

Message + Cover Image

1 2 3 4 5 6 1 2 3 4 5 6

1 1 2 7 4 5 6 1 17 2 7 18 5 25

2 7 4 9 10 11 12 2 7 21 9 10 11 12

3 13 14 15 11 17 11 3 13 14 0 11 12 11

4 19 14 21 22 23 24 4 19 0 2 23 23 24

Pixel Image ASCII

row col Value Value

1 3 = 4 + 65 = H

2 2 = 21 + 65 = V

3 4 = 11 + 65 = L

3 6 = 11 + 65 = L

4 2 = 0 + 65 = A

Steganographic Image Steganalysis Image

Modulus 25

Pixel

Stego + Neutralization

Cover ImageASCII

Image/Video Formats

The bandwidth that image and video can support are large. A

standard definition (480 x 720) true color image can store

1,036,800 characters (~2 Books).

A high definition (1080 x 1920)

true color image can store enough

information to fill ~13 books. A

4K true color image can hold

enough information to fill ~50

books. And a 4K true color 60 Hz

video and store enough

information to fill ~180,000 books.

Since Steganalysis will attempt to neutralize the

effectiveness of the steganographic image via manipulating

the image pixels, quality image metrics were required to

determine the video quality impacted by this steganalysis

process. To measure degradation between the

steganographic image and resulting steganalysis image, a

class of quality assessment metrics called full reference

(FR) were considered. Full reference metrics perform

distortion measurements having full access to the original

image (i.e. Steganographic Image). Two FR metrics were

selected: Mean Square Error (MSE) and Structural

Similarity (SSIM). MSE is the squared average difference

between Steganographic Image and Sanitized Steganalysis

Image pixels. This metric is simple to calculate but does not

align well with the human perception. The MSE metric was

selected to support Video Analytics/ Computer Vision

requirements which allows computers to monitor the

prison’s front gate.9 SSIM was selected to allow for human

observation which combines local image structure,

luminance, and contrast into a single local quality score.

Structures are patterns of pixel intensities, especially among

neighboring pixels, after normalizing for luminance and

contrast because the human visual system is good at

perceiving structure, the SSIM quality metric agrees more

closely with the subjective quality scoring.10

III. THE SCENARIO/BENCHMARK

This paper developed a challenging scenario to help explore

different approaches and domains using the modified

prisoner’s problem described above. This scenario used a

1,200 character (i.e. 9,600 bits) escape message that is

embedded via steganography into the Prison Image/Video

(assume one 480 x 720 pixel 8 bit grayscale frame),

comprised of 2,764,800 bits. The required message

bandwidth only requires 0.347% (9,600 Escape Message

bits / 2,764,800 Cover Prison Image bits). The

steganographic image created by Alice cannot have any

noticeable artifacts or the Warden will know that the prison

surveillance image has been altered. In addition, any

process to eliminate or effectively neutralize the message

(via steganalysis cannot impact the MSE greater than 10 nor

reduce the SSIM less than 0.9800. Both the MSE and SSIM

are measured between the steganographic image and the

steganalysis image. If the steganalysis image corrupts the

image by more than the MSE of 10, the Video

Analytics/Computer Vision capabilities will not be

effective (i.e. prisoners walk out the front gate without

being detected by Computer Vision algorithms). Similarly,

if the SSIM falls below 0.9800, then the human prison

guards monitoring the prison gate will not be effective (i.e.

prisoners walk out the front gate without being detected by

the prison guards). In addition, only 25% of the escape

message has to be received by Bob to be successful (i.e. if

Bob receives at least 25% of the message, he will

understand the message and escape).

The steganographic process takes the 1,200 character

message and converts each character to ASCII, then sixty-

five was subtracted) and then converted to Binary (e.g. H -

> 72 -> 7 -> 0000111) resulting in a tot al o f 9,600 bits.

Each message bit is then embedded into the Cover image

byte with no more than one message bit embedded into each

message byte.

Example #1: One Message Bit is embedded into every Least

Significant Bit (LSB) of each Cover Byte. With a total of

345,600 cover (Prison Image) bytes, the 9,600 bit message is

placed in the cover image LSB thirty-six times as illustrated

in the diagram below.

The cover image used throughout this paper is the

image/video of the prison front gate seen below.

Co

ver

Imag

e P

ixel

(1,1

)

Co

ver

Imag

e P

ixel

(1,2

)

Co

ver

Imag

e P

ixel

(1

,3)

Co

ver

Imag

e P

ixel

(1

,4)

Co

ver

Imag

e P

ixel

(1

,5)

Co

ver

Imag

e P

ixel

(1

,6)

Co

ver

Imag

e P

ixel

(1

,7)

Co

ver

Imag

e P

ixel

(1

,8)

Co

ver

Imag

e P

ixel

(1

,9)

Co

ver

Imag

e P

ixel

(1

,10

)

Co

ver

Imag

e P

ixel

(1

,11

)

Co

ver

Imag

e P

ixel

(1

,12

)

Co

ver

Imag

e P

ixel

(1

,13

)

Co

ver

Imag

e P

ixel

(1

,14

)

Co

ver

Imag

e P

ixel

(1

,15

)

Co

ver

Imag

e P

ixel

(1

,16

)

Co

ver

Imag

e P

ixel

(1

,17

)

Bit #8 C C C C C C C C C C C C C C C C C

Bit #7 C C C C C C C C C C C C C C C C C

Bit #6 C C C C C C C C C C C C C C C C C

Bit #5 C C C C C C C C C C C C C C C C C

Bit #4 C C C C C C C C C C C C C C C C C

Bit #3 C C C C C C C C C C C C C C C C C

Bit #2 C C C C C C C C C C C C C C C C C

Bit #1 M M M M M M M M M M M M M M M M M

1st 2nd 3rd 4th 5th 6th 7th 8th 1st 2nd 3rd 4th 5th 6th 7th 8th 1st

…..

C = Cover Image Bit (Prison)

M = Message Bit

Message 1st Character Message 2nd Character

True Color Bandwidth

The cover “Prison Gate” image

The Cover Image Histogram provides a frequency

distribution of all the 256 grayscale colors in the image (0 =

Black and 255 = White). It can be noted the histogram

distribution is fairly smooth between grayscale color and the

next (i.e. no discontinuities are seen). For example, the

number pixels that have a grayscale color of 100 within the

prison image is 2,632 out of a total of 345,600 pixels.

The image quality metrics between the original and original

and original and steganographic is provided below. Clearly,

the MSE is zero and SSIM is one when comparing the same

cover image to itself. Even after injecting the message bits

into the LSB of the cover image, the image quality remains

very good (MSE: 0.5026 < 10 and SSIM: 0.9972 > 0.9800).

When embedding the message into the image’s Least

Significant Bit, the MSE has only affected approximately

50% of the image bytes. This makes sense, in that on

average the embedded message only changes 50% of the

image bytes by at most one (e.g. Message Bit = 0 and Image

Byte = 7, Steganographic Image Byte = 6). In addition, the

SSIM has not been impacted much.

However, there is a noticeable structural deviation in the

Steganographic (Cover Image + Embedded Message)

image histogram as compared to the Cover image

histogram.

Example #2: One Message Bit is embedded into every Third

Significant Bit (LSB) of each Cover Image byte. With a

total of 345,600 cover image bytes, the 9,600 bit message

can be placed in the cover image LSB thirty-six times.

The steganographic image does not have any perceptible

image degradations.

17

Co

ver

Imag

e P

ixel

(1,1

)

Co

ver

Imag

e P

ixel

(1,2

)

Co

ver

Imag

e P

ixel

(1

,3)

Co

ver

Imag

e P

ixel

(1

,4)

Co

ver

Imag

e P

ixel

(1

,5)

Co

ver

Imag

e P

ixel

(1

,6)

Co

ver

Imag

e P

ixel

(1

,7)

Co

ver

Imag

e P

ixel

(1

,8)

Co

ver

Imag

e P

ixel

(1

,9)

Co

ver

Imag

e P

ixel

(1

,10

)

Co

ver

Imag

e P

ixel

(1

,11

)

Co

ver

Imag

e P

ixel

(1

,12

)

Co

ver

Imag

e P

ixel

(1

,13

)

Co

ver

Imag

e P

ixel

(1

,14

)

Co

ver

Imag

e P

ixel

(1

,15

)

Co

ver

Imag

e P

ixel

(1

,16

)

Co

ver

Imag

e P

ixel

(1

,17

)

Bit #8 C C C C C C C C C C C C C C C C C

Bit #7 C C C C C C C C C C C C C C C C C

Bit #6 C C C C C C C C C C C C C C C C C

Bit #=5 C C C C C C C C C C C C C C C C C

Bit #4 C C C C C C C C C C C C C C C C C

Bit #3 C C C C C C C C C C C C C C C C C

Bit #2 C C C C C C C C C C C C C C C C C

Bit #1 M M M M M M M M M M M M M M M M M

1st 2nd 3rd 4th 5th 6th 7th 8th 1st 2nd 3rd 4th 5th 6th 7th 8th 1st

…..

C = Cover Image Bit (Prison)

M = Message Bit

Message 1st Character Message 2nd Character

Steganographic Image (Prison Front Gate)

With the message bits being inserted in to the 3rd bit of every

cover image byte, the image quality is starting to degrade.

Although the MSE is less than 10 (MSE: 8.06 < 10), the

SSIM has fallen to below 0.9800 (SSIM: 0.9669 > 0.9800),

which impacts human observability. So this encoding fails

the Scenario outlined above (via SSIM > 0.9800

constraint).

This steganalysis image distortion is noticeable. See red

circle.

Steganographic Image 3rd LSB (Prison Front Gate)

As in Example #1, there are noticeable structural deviations

in the steganographic image histogram as compare to the

cover image histogram, which is located in the upper right

hand corner of the histogram.

Example 3: One Message Bit is embedded into every

Fourth Significant Bit (LSB) of each Cover Image byte.

With a total of 345,600 cover (Prison Image) bytes, the

9,600 bit message can be placed in the cover image LSB

thirty-six times.

With the message bits being inserted into every 4th bit for the

cover image, the image quality has degraded. Both the MSE

is above 10 (MSE: 35.98 > 10), and the SSIM has fallen to

below 0.9800 (SSIM: 0.8945 < 0.9800), which both impacts

video analytics/computer vision and human observability.

So this encoding fails the Scenario outlined above.

3rd Bit 36X Message

Original/Original Image

Steganography Image/Original

MSE (0.9800) 1.0000 0.9669

Co

ver

Imag

e P

ixel

(1,1

)

Co

ver

Imag

e P

ixel

(1,2

)

Co

ver

Imag

e P

ixel

(1

,3)

Co

ver

Imag

e P

ixel

(1

,4)

Co

ver

Imag

e P

ixel

(1

,5)

Co

ver

Imag

e P

ixel

(1

,6)

Co

ver

Imag

e P

ixel

(1

,7)

Co

ver

Imag

e P

ixel

(1

,8)

Co

ver

Imag

e P

ixel

(1

,9)

Co

ver

Imag

e P

ixel

(1

,10

)

Co

ver

Imag

e P

ixel

(1

,11

)

Co

ver

Imag

e P

ixel

(1

,12

)

Co

ver

Imag

e P

ixel

(1

,13

)

Co

ver

Imag

e P

ixel

(1

,14

)

Co

ver

Imag

e P

ixel

(1

,15

)

Co

ver

Imag

e P

ixel

(1

,16

)

Co

ver

Imag

e P

ixel

(1

,17

)

Bit #8 C C C C C C C C C C C C C C C C C

Bit #7 C C C C C C C C C C C C C C C C C

Bit #6 C C C C C C C C C C C C C C C C C

Bit #=5 C C C C C C C C C C C C C C C C C

Bit #4 C C C C C C C C C C C C C C C C C

Bit #3 C C C C C C C C C C C C C C C C C

Bit #2 C C C C C C C C C C C C C C C C C

Bit #1 M M M M M M M M M M M M M M M M M

1st 2nd 3rd 4th 5th 6th 7th 8th 1st 2nd 3rd 4th 5th 6th 7th 8th 1st

…..

C = Cover Image Bit (Prison)

M = Message Bit

Message 1st Character Message 2nd Character

4th Bit 36X Message

Original/Original Image

Steganography Image/Original

MSE (0.9800) 1.0000 0.8945

Steganographic Image 4th LSB (Prison Front Gate)

As in Example #1 and #2, there are structural deviations of the

steganographic image histogram as compare to the cover

image histogram (located in the upper left hand corner).

Because of the above steganographic image histograms

structural deviations, additional steganographic images were

investigated to see if the steganographic image histogram

artifacts discovered above are image dependent or image

independent.

A “Missile Firing” grayscale image was LSB steganographically embedded using the same process

discussed above. The structural histogram anomalies are

similar to the Prison Gate steganographic image.

A “German Tiger Tank” grayscale image was LSB

steganographically embedded using the same process discussed above. The structural histogram anomalies are

similar to the Prison Gate steganographic image.

A “Storm” grayscale image was LSB steganographically

embedded using the same process discussed above. The

structural histogram anomalies are similar to the Prison Gate

steganographic image.

A “Scenic” grayscale image was 2nd LSB steganographically

embedded using the same process discussed above. The

structural histogram anomalies are similar to the Prison Gate steganographic image.

To determine the cause of this anomaly, a careful review

of the steganographic histograms was performed. It can

be seen that for the steganographic image that placed the

message into the Least Significant Bit (LSB), the

steganographic histogram is significantly different than

the original cover image. In the steganographic image

histogram there is significantly more even numbered

pixels than odd.

When the message is placed in the 2nd significant bit of

the cover image, the x+1 and x+2 values are more

numerous than the x+3 and x+4 values (e.g.

0,1,4,5,8,9,12,13… have ~3x the 2,3,6,7,10,11…

numbers.

This pattern is repeated when placing the steganographic

message in both the 3rd and 4th significant bit of the

image. The histogram gap is associated with the bit

placement. 1st significant bit has a gap of 2^0=1 bit gap,

2nd bit has a gap of 2^1 = 2 bit gap, 3rd bit produces a

2^2=4 bit gap and a 4th bit placement yields a 2^3 = 8 bit

gap.

Stego-Image Histogram1st Bit Message Placement

Stego-Image Histogram2nd Bit Message

Cover Image Histogram

Stego-Image Histogram1st Bit Message Placement

1 Bit Width20 = 1

Stego-Image Histogram2nd Bit Message

2 Bit Width21 = 2

Stego-Image Histogram3rd Bit Message

3rd Bit Width22 = 4

Stego-Image Histogram4th Bit Message

4th Bit Width23 = 8

This led to an investigation of the steganographic message

encoding. The message was ASCII encoded and then 65 was

subtracted to provide an alphabetic range of zero to twenty-

five (0…25). This was done to support various encryption

techniques (Ceaser Cipher, Affine Cipher, One-Time-Pad).

This encoding binary scheme injects 271% more 0’s than 1’s

(i.e. 56 Ones to 152 Zeros). The ASCII requires 8 bits to

support these 26 letters. If the encoding is modified to

support Radix-6 the difference between there are 178% more

0’s than 1’s (i.e. 56 Ones to 100 Zeros). If one reduces the

encoding to 5 bits, then there are 132% more 0’s than 1’s (i.e.

56 Ones to 74 Zeros).

The MSE is significantly reduced using 5 bit encoding.

The MSE between the cover image and the ASCII

encoding is 2,116,400. The MSE can be reduced to

569,280 using the Radix-6 encoding. The MSE can be

reduced further to 180,270 using 5 bit encoding scheme.

Encryption methods helped improve MSE.

The first Encryption Algorithm investigated was the

Affine. ci = f(mi) = ((k1 x mi) + k2 )mod 26

gcd (mi, 26) = 1 mi = f

-1(ci) = (k1-1 x (ci – k2 ))mod

26 k1 = 17, k2 = 8, k1-1 = 23

The Affine encryption improved the MSE from 180,270

to 88,373.

The Affine Frequency Distribution is provide below:

The one-time-pad frequency histogram “flattened” out

the frequency distribution.

There was still appears some minor differences between

the cover image and the steganographic image using

Radix 5 and One-Time-Pad encryption. But the MSE

1 0 1.00 0.00 1 0 1.00 0.00

65 A 0 0 0 6 7.25% 0.00 0.44 0 5 7.25% 0.00 0.36

66 B 1 1 1 5 1.25% 0.07 0.36 1 4 1.25% 0.07 0.29

67 C 2 10 1 5 3.50% 0.07 0.36 1 4 3.50% 0.07 0.29

68 D 3 11 2 4 4.25% 0.15 0.29 2 3 4.25% 0.15 0.22

69 E 4 100 1 5 12.75% 0.07 0.36 1 4 12.75% 0.07 0.29

70 F 5 101 2 4 3.00% 0.15 0.29 2 3 3.00% 0.15 0.22

71 G 6 110 1 5 2.00% 0.07 0.36 1 4 2.00% 0.07 0.29

72 H 7 111 3 3 3.50% 0.22 0.22 3 2 3.50% 0.22 0.15

73 I 8 1000 1 5 7.75% 0.07 0.36 1 4 7.75% 0.07 0.29

74 J 9 1001 2 4 0.25% 0.15 0.29 2 3 0.25% 0.15 0.22

75 K 10 1010 2 4 0.50% 0.15 0.29 2 3 0.50% 0.15 0.22

76 L 11 1011 3 3 3.75% 0.22 0.22 3 2 3.75% 0.22 0.15

77 M 12 1100 2 4 2.75% 0.15 0.29 2 3 2.75% 0.15 0.22

78 N 13 1101 3 3 7.75% 0.22 0.22 3 2 7.75% 0.22 0.15

79 O 14 1110 3 3 7.50% 0.22 0.22 3 2 7.50% 0.22 0.15

80 P 15 1111 4 2 2.75% 0.29 0.15 4 1 2.75% 0.29 0.07

81 Q 16 10000 1 5 0.01% 0.07 0.36 1 4 0.01% 0.07 0.29

82 R 17 10001 2 4 8.50% 0.15 0.29 2 3 8.50% 0.15 0.22

83 S 18 10010 2 4 6.00% 0.15 0.29 2 3 6.00% 0.15 0.22

84 T 19 10011 3 3 9.25% 0.22 0.22 3 2 9.25% 0.22 0.15

85 U 20 10100 2 4 3.00% 0.15 0.29 2 3 3.00% 0.15 0.22

86 V 21 10101 3 3 1.50% 0.22 0.22 3 2 1.50% 0.22 0.15

87 W 22 10110 3 3 1.50% 0.22 0.22 3 2 1.50% 0.22 0.15

88 X 23 10111 4 2 0.50% 0.29 0.15 4 1 0.50% 0.29 0.07

89 Y 24 11000 2 4 2.25% 0.15 0.29 2 3 2.25% 0.15 0.22

90 Z 25 11001 3 3 0.25% 0.22 0.22 3 2 0.25% 0.22 0.15

56 100 4.06 7.25 56 74 4.06 5.37

Radix 6 Radix 5ASCII

Mean Square Error (MSE)2,116,400 ASCII Coding569,280 Radix6 Coding180,270 Radix5 Coding88,373 Radix5 & Affine Cipher13,388 Radix5 & One Time Pad

Cover Image Steganographic ImageASCII Coded2,116,400 MSE

Steganographic ImageRadix-6 Coded569,280 MSE

Steganographic ImageRadix-5 Coded180,270 MSE

Steganographic ImageRadix CodedOne-Time Pad Encrypted13,388 MSE

has been reduced from 2,116,400 to 13,388.

IV. THE CHALLENGE STEGANOGRAPHY VERSUS STEGANALYSIS – WHO WINS?

Alice will steganographically embed a 1,200 character

message (9,600 bits) into a standard definition (480 x 720

pixel) grayscale image. Due to the steganographic image

histogram vulnerabilities discovered above, all

steganographic images will only embedded the message once

(9,600 Message bits into 2,764,800 Cover Image bits). Given

this new steganographic approach, will Bob receive the

message (>25% of the 1200 characters) and escape or will

the Warden effectively neutralize the steganographic image

while maintaining quality surveillance image/video (MSE

0.9800). Let’s see what happens.

Scenario #1: Alice via steganography places the message

once randomly in the Least Significant Bit (LSB).

The steganographic image appears normal.

Unfortunately, if Kerchkoff’s Principle is followed and the

steganographic algorithm is known, the logical steganalysis

neutralization approach would be to set every LSB to “0”.

This meets the criteria of the constraints. The steganalysis

image’s MSE is less than 10 (0.4961) and the SSIM is greatly

than .9800 (0.9986). This approach effectively eliminates the

steganographic message by only allowing 8.8% of the

message to get through to Bob.

Scenario #2: Alice places the message randomly into the

Cover Image (e.g. Message – Character 1 2nd Bit is

randomly placed the Cover Image Pixel (1,4) in the 8th bit.)

The steganographic image has a noticeable amount of

artifacts that was created by the steganographic process

and is easy to identify. Therefore this is not an

acceptable approach.

Cover ImageC

ove

r Im

age

Pix

el(1

,1)

Co

ver

Imag

e P

ixel

(1,2

)

Co

ver

Imag

e P

ixel

(1

,3)

Co

ver

Imag

e P

ixel

(1

,4)

Co

ver

Imag

e P

ixel

(1

,5)

Co

ver

Imag

e P

ixel

(1

,6)

Co

ver

Imag

e P

ixel

(1

,7)

Co

ver

Imag

e P

ixel

(1

,8)

Co

ver

Imag

e P

ixel

(1

,9)

Co

ver

Imag

e P

ixel

(1

,10

)

Co

ver

Imag

e P

ixel

(1

,11

)

Co

ver

Imag

e P

ixel

(1

,12

)

Co

ver

Imag

e P

ixel

(1

,13

)

Co

ver

Imag

e P

ixel

(1

,14

)

Co

ver

Imag

e P

ixel

(1

,15

)

Co

ver

Imag

e P

ixel

(1

,16

)

Co

ver

Imag

e P

ixel

(1

,17

)

Bit #8 C C C C C C C C C C C C C C C C C

Bit #7 C C C C C C C C C C C C C C C M C

Bit #6 C C C C C C C C C C C C C C C C C

Bit #=5 C C C C C C C C C C C C C C C C C

Bit #4 C C C C C C C C C C C C C C C C C

Bit #3 C C C C C C C C C C C C C C C C C

Bit #2 C C C C C C C C C C C C C C C C C

Bit #1 M C C M C C C C M M C C C C C M C

1st 2nd 3rd 4th 5th

C = Cover Image Bit (Prison)

M = Message Bit

Message 1st Character

Random 1st Bit One Message

Original Image/Steganographic

Steganographic/ Steganaylsis

MSE (0.9800) 0.9999 0.9986

Correct Characters (out of 1200)>25%

8.8%

Co

ver

Imag

e P

ixel

(1,1

)

Co

ver

Imag

e P

ixel

(1,2

)

Co

ver

Imag

e P

ixel

(1

,3)

Co

ver

Imag

e P

ixel

(1

,4)

Co

ver

Imag

e P

ixel

(1

,5)

Co

ver

Imag

e P

ixel

(1

,6)

Co

ver

Imag

e P

ixel

(1

,7)

Co

ver

Imag

e P

ixel

(1

,8)

Co

ver

Imag

e P

ixel

(1

,9)

Co

ver

Imag

e P

ixel

(1

,10

)

Co

ver

Imag

e P

ixel

(1

,11

)

Co

ver

Imag

e P

ixel

(1

,12

)

Co

ver

Imag

e P

ixel

(1

,13

)

Co

ver

Imag

e P

ixel

(1

,14

)

Co

ver

Imag

e P

ixel

(1

,15

)

Co

ver

Imag

e P

ixel

(1

,16

)

Co

ver

Imag

e P

ixel

(1

,17

)Bit #8 C C C M C C C C C C C C C C C C C

Bit #7 C C C C C C C C C C C C C C C M C

Bit #6 C C C C C C C C C C C C C C C C C

Bit #=5 C C C C C C C C M C C C C C C C C

Bit #4 C C C C C C C C C C C C C C C C C

Bit #3 C C C C C C C C C M C C C C C C C

Bit #2 C C C C C C C C C C C C C C C C C

Bit #1 M C C C C C C C C C C C C C C C C

1st 2nd 3rd 4th 5th

C = Cover Image Bit (Prison)

M = Message Bit

Message 1st Character

Given Kerchkoff’s Principle, steganalysis is performed by

randomly embedding one bit into each byte on the entire

steganographic image. Unfortunately, this appreciably

degrades the steganalysis image.

It appears that placing the Steganographic Message

randomly in any bit of the Cover Image successfully allows

the message to be recovered (32%, requirement is 25%) and

the steganalysis neutralization greatly exceeds the thresholds

(i.e. MSE = 2,711 >10 and SSIM = 0.1525 < 0.9800).

Unfortunately for Alice, the steganographic image did not

pass the undetectability requirement since artifacts were

clearly seen in the image.

Scenario #3: Given that scenario #2 steganographic image

has noticeable amounts of artifacts, Alice places the

message randomly in in any of the cover image’s first four

least significant bits (e.g. Message – Character 1 2nd Bit is

randomly placed the Cover Image Pixel (1,4) in the 8th bit.

Even knowing the Steganographic approach via

Kerchkoff’s Principle, and randomly injecting noise in the

lowest 4 bits is can eliminate the message without

noticeable distorted the image (i.e. MSE > 10 and SSIM <

0.9800).

To effectively reduce the message to less than 25%, requires

a MSE of 14.2 (>10) and SSIM of 0.9143( 0.9800).

V. NEW APPROACH

To effectively neutralize the steganographic message a

new steganalysis algorithm was required to replace the

random noise injection approach. A sliding one-

dimensional filter was created. This sliding filter takes the

nth pixel and multiplies it by a filter value A and adds the

immediately pixel neighbors (n-1 and n+1) to the left and

right and multiples by another smaller filter value B to

produce the new value for that pixels. Boundary pixels

remain the same.

Filtered_Pixel (x,y) =

Steganographic Image

Steganalysis Image

Random1st–8th

One MessageOriginal Image/Steganographic

Steganographic/ Steganaylsis

MSE (0.9800) 0.9057 0.1525

Correct Characters (out of 1200) 32.0%

Co

ver

Imag

e P

ixel

(1,1

)

Co

ver

Imag

e P

ixel

(1,2

)

Co

ver

Imag

e P

ixel

(1

,3)

Co

ver

Imag

e P

ixel

(1

,4)

Co

ver

Imag

e P

ixel

(1

,5)

Co

ver

Imag

e P

ixel

(1

,6)

Co

ver

Imag

e P

ixel

(1

,7)

Co

ver

Imag

e P

ixel

(1

,8)

Co

ver

Imag

e P

ixel

(1

,9)

Co

ver

Imag

e P

ixel

(1

,10

)

Co

ver

Imag

e P

ixel

(1

,11

)

Co

ver

Imag

e P

ixel

(1

,12

)

Co

ver

Imag

e P

ixel

(1

,13

)

Co

ver

Imag

e P

ixel

(1

,14

)

Co

ver

Imag

e P

ixel

(1

,15

)

Co

ver

Imag

e P

ixel

(1

,16

)

Co

ver

Imag

e P

ixel

(1

,17

)

Bit #8 C C C C C C C C C C C C C C C C C

Bit #7 C C C C C C C C C C C C C C C C C

Bit #6 C C C C C C C C C C C C C C C C C

Bit #=5 C C C C C C C C C C C C C C C C C

Bit #4 C C C M C C C C C C C C C C C C C

Bit #3 C C C C C C C C C M C C C C C C C

Bit #2 C C C C C C C C M C C C C C C C C

Bit #1 M C C C C C C C C C C C C C C M C

1st 2nd 3rd 4th 5th

C = Cover Image Bit (Prison)

M = Message Bit

Message 1st Character

Random 1-4 Bit/1 MessageRand1-8 Bit/every 1.5 Byte

Original Image/Steganographic

Steganographic/ Steganaylsis

MSE (0.9800) 0.9973 0.9143

Correct Characters (out of 1200) 21.8%

Pixel(x, y-1)* Filter_Value_B +

Pixel (x,y) * Filter_Value_A +

Pixel(x, y+1)* Filter_Value_B

1 = A + 2*B

The example below uses Filter_Value_A = 0.8 and

Filter_Value_B = 0.1

This steganalysis filter concept was very effective. The

percentage of successful message was reduced to 10.4%

and maintained image quality with MSE(5.15) < 10 and

SSIM(0.9930) > 0.9800. Even though the message only

required less than 0.347% of the cover image bandwidth,

no steganographic algorithm was found that could place a

1,200 character message into a grayscale standard

definition image and successfully recover at least 25% of

the message or required additional steganalysis that

resulted in reduce video quality.

VI. FREQUENCY DOMAIN

There are steganographic approaches that do not directly

modifying the images’ pixels. One approach is to exploit

the image’s frequency domain. An example is the Joint

Photographic Experts Group (JPEG). In 1992, JPEG

became an international standard for compressing digital

still images. There are four basic steps in the JPEG

algorithm - preprocess, transformation, quantization, and

coding.11 Starting with a grayscale image, step one is to

subtract 127 from each image pixel value and then

partition the image into 8 x 8 pixel blocks. Since we are

using 480 by 720 standard definition images, this equates

to 5,400 blocks (480 x 720 / (8 x 8)) blocks. This

preprocessing has done nothing that will make the coding

portion of the algorithm more effective. The

transformation step is the key to increasing the coder's

effectiveness. The JPEG image compression standard

relies on the Discrete Cosine Transformation (DCT) to

transform the image. The DCT is a product C = U*B*U^T

where B is an 8 x 8 block of the preprocessed image and

U is a special 8 x 8 matrix (i.e. DCT matrix). The DCT

tends to push most of the high intensity information

(larger values) in the 8 x 8 block to the upper left-hand

corner of the matrix C with the remaining values in C

taking on relatively small values. The DCT is applied to

each 8 x 8 block. The DCT specific values are provide

below:

The next step in the JPEG algorithm is the quantization

step. The JPEG algorithm first divides each element by

the “Z” matrix and then rounds the result to produce

integers. Elements near zero will be converted to zero.

Quantization makes the JPEG algorithm an example of

lossy compression. The DCT (C = U*B*U^T) step is

completely invertible. It turns out we can recover B by

the computation B = U^T*C*U. However, converting

small values to 0 and rounding all quantized values are

not reversible steps and will forever lose the ability to

recover the original image. Quantization is performed in

order to obtain integer values and to convert a large

number of the values to 0. The “Z” quantization matrix

is provided below:

The last step in the JPEG process is to code the

transformed and quantized image. The regular JPEG

standard uses an advanced version of Huffman coding.

Below is an example of an 8 x 8 image block that is

Preprocessed, Transformed and then Quantized. The

upper left is an 8 bit grayscale 8 x 8 matrix from a

steganographic image. Preprocessing subtracts 127 from

this matrix (upper right). The lower left is the

Transformed matrix C = U*B*U^T where B is an 8 x 8

block from the preprocessed image and U is a special 8 x

8 matrix (i.e. DCT). The lower right block shows the

quantized matrix, with 50% of the values equal to zero.

0.1 0.8 0.1

1 2 3 4 5 6 7 8 9 10

1 250 100 50 75 80 98 5 100 59 1

2 200 150 100 75 25 0 50 100 200 2

3 150 50 75 0 50 100 200 100 0 3

1 2 3 4 5 6 7 8 9 10

1 250 110 57.5 73 81.3 86.9 23.8 86.4 57.3 1

2 200 150 102.5 72.5 27.5 7.5 50 105 170.2 2

3 150 62.5 65 12.5 50 105 180 100 10.3 3

Random 1-4 Bit/1 MessageFilter 0.1-0.8-0.1

Original Image/Steganographic

Steganographic/ Steganaylsis

MSE (0.9800) 0.9973 0.9930

Correct Characters (out of 1200) >25% 10.4%

The 8 bit grayscale prison image size is 345,600 bytes

(480 x 720), using the JPEG algorithms this image size is

compressed to 77,525 bytes. This equates to a 78%

reduction in memory size. Alice only requires 0.347%

of the 8 bit grayscale image to embedded the 9,600 bit

message. Unfortunately, Alice needs 1.55% of the JPEG

image bandwidth (1,200 bytes / 77,525 bytes).

Analysis of both frequency domain steganography and

steganalysis was problematic in supporting the revised

prisoner’s problem. The matlab code developed for this

effort took the grayscale BMP image and converted it to

JPEG image using the approach described above. The

image quality between the BMP and JPEG images was

large with visual image distortion visible.12

Placing the 9,600 bit message into the JPEG DCT matrix

caused considerable distortion. Just placing the 9,600 bit

message once into the Least Significant Bit (LSB)

77,525 byte JPEG matrix caused addition visual image

degradations. With the MSE being driven to 207 and

SSIM to 0.6440.

Placing the 9,600 bit message into the 2nd Least

Significant Bit (LSB) of the 77,525 byte JPEG matrix

caused significant visual image degradations. With the

MSE being driven to 489 and SSIM to 0.5224.

Placing the 9,600 bit message into the 4th Least

Significant Bit (LSB) of the 77,525 byte JPEG matrix

caused bad visual image degradations. With the MSE

being driven to 3,309 and SSIM to 0.2431.

Looking at the JPEG histogram, one can see that the

BMP image is uneven, vice the JPEG histogram is

smother.

1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

1 4 8 12 26 20 24 28 32 1 -123 -119 -115 -101 -107 -103 -99 -95

2 36 40 44 48 52 56 60 64 2 -91 -87 -83 -79 -75 -71 -67 -63

3 44 8 12 16 20 24 28 32 3 -83 -119 -115 -111 -107 -103 -99 -95

4 36 40 44 48 52 56 60 256 4 -91 -87 -83 -79 -75 -71 -67 129

5 4 8 12 16 20 24 28 208 5 -123 -119 -115 -111 -107 -103 -99 81

6 36 40 44 48 52 56 60 224 6 -91 -87 -83 -79 -75 -71 -67 97

7 164 8 12 16 20 24 28 240 7 37 -119 -115 -111 -107 -103 -99 113

8 256 200 224 256 244 248 252 256 8 129 73 97 129 117 121 125 129

1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

1 -435.0 -166.3 153.5 -80.7 128.0 -44.7 67.4 -19.5 1 -27 -15 15 -5 5 -1 1 0

2 -354.7 36.9 -86.2 22.8 -80.6 6.3 -41.0 7.8 2 -30 3 -6 1 -3 0 -1 0

3 202.1 89.5 -62.9 83.9 -34.4 64.4 -21.0 17.4 3 14 7 -4 3 -1 1 0 0

4 -221.1 -47.7 41.0 -47.6 19.0 -39.3 12.5 -9.1 4 -16 -3 2 -2 0 0 0 0

5 167.0 -35.2 -32.1 -23.7 -14.0 -9.4 -9.5 -7.3 5 9 -2 -1 0 0 0 0 0

6 -129.6 -4.9 80.6 -8.9 53.4 -11.0 30.4 -0.7 6 -5 0 1 0 1 0 0 0

7 51.6 18.0 -68.0 18.6 -50.1 15.9 -26.1 3.4 7 1 0 -1 0 0 0 0 0

8 -153.2 38.2 20.4 30.7 12.7 18.7 7.4 7.7 8 -2 0 0 0 0 0 0 0

Cover Image - Preprocess

Cover Image - QuantiizationCover Image - Transformation

Cover Image - 8 x 8

PDF Image

MSE = 81.0SSIM = 0.8019

PDF Image – 1st Bit

MSE = 207SSIM = 0.6440

PDF Image – 2nd Bit

MSE = 489SSIM = 0.5244

MSE = 3,309SSIM = 0.2431

PDF Image – 4th Bit

Embedding messages into the JPEG DCT matrix has the

effect of spreading the histogram. If messages are

embedded into the 4th significant bit, the histogram

flattens out.

To validate these points, the 9,600 bit message was

embedded one time randomly into the Least Significant

Bit (LSB). Steganalysis embedded randomly into every

6th LSB of the steganographic image. Both the

steganographic image and the steganalysis image has

significant distortions.

The image quality for both steganographic image and

steganalysis image are poor with the steganographic

image’s MSE being 97 and SSIM being 0.7720. The

steganalysis image did not fare much better.

VII. TIME DOMAIN

Up to this point, we have only discussed image

steganography (i.e. one frame of the video). As shown

above, it is very difficult to embedded a large message in

an uncontrolled image (i.e. the steganographer has no

control over the selection of the cover image, in this case

the prison surveillance video). As seen above, Alice, the

steganographer embedded a 1200 character message into

a standard depth image randomly over the first 4 image

bits. The steganalysis performed a simple 3 step filter

and successful neutralized the message (only

allowed10.4% of the message to make it to Bob) and

minimized impact to video quality (MSE 0.9800 (0.9973). Therefore the steganalysis

had the advantage in neutralizing the message using a

three element filter. But, what if the steganographer

could leverage video, what that change?

In this scenario, if the steganographer replaced the 9,600

bit full message with only 25 characters of the message

for each frame/image, repeated 100 times in the

image/frame so that after 48 frames (~1 video second, 25

characters x 48 frames = 1,200 characters) the complete

message is transmitted. Using the same steganalysis

approach, 92% of the message is retrieved by Bob.

Increasing the filter coefficients increases MSE and

decreases SSIM to unacceptable levels. In this case a

0.15-.7-0.15 filter allows 56% of the message to be

received and the MSE exceeds 10 (11.31), but the SSIM

is above the acceptable level of 0.9800 (0.9848).

VIII. CONCLUSION

Lesson learned from these exercises:

1) Control and selection of the cover image is

important. Not choosing the cover image impacts the

performance of the steganography image. If the

steganographer could select the cover image, they could

closely match the cover image histogram characteristics

with that of the embedded message and select an effect

encoding steganographic algorithm. A matlab code could

be written that could check 1000s of cover images to

determine which is the best cover image that reduces the

MSE to less than 10, maintains the SSIM close to 1.000

and minimizes steganographic image histogram artifacts

(as seen above).

2) Message encoding is very important and needs to

balance 0’s and 1’s. As was seen above an unbalanced

encoding schemes will reduce the effective

steganographic bandwidth. Even a steganographic

algorithm that embeds ~30% more zeros than ones into

the cover image can be detected via a histogram analysis.

3) Larger image bandwidth favors steganography, by

providing more places to hide. Embedding 9,600 bits

into a cover image with 2,764,800 bits even using

balanced encoding allowed effective hiding capability.

Applying the same algorithm on High Definition 1080 x

1920 would allow 7,200 characters to be hidden or a 4K

3840 x 2160 image would allow 28,800 message to be

hidden with the same MSE and SSIM results.

Cover BMP Image Cover PDF ImageSmooth Distribution

Stego ImageLSB

Stego Image2nd LSB

Stego Image4th LSB

Steganographic PDF Image Steganalysis PDF Image

Random 1-4 Bit/1 MessageFilter 0.1-0.8-0.1

Original Image/Steganographic

Steganographic/ Steganaylsis

MSE (0.9800) 0.9973 0.9930

Correct Characters (out of 1200) 10.4%

Correct Characters (out of 25) repeated 100x 92%

Random 1-4 Bit/1 MessageFilter 0.15-0.7-0.15

Original Image/Steganographic

Steganographic/ Steganaylsis

MSE (0.9800) 0.9973 0.9848

Correct Characters (out of 1200) 10.4%

Correct Characters (out of 25) repeated 100x 56%

4) Cleaner cover image favors steganalysis, since

steganography exploits image/video noise.

5) Video favors steganography, by spreading the

message across the video. The limitation in sending

concealed long messages in one standard definition

image frame is overcome by using larger formats (HD,

4K) and video. This was seen by send above by

spreading the message across multiple video frames and

only sending 200 message bits vice 9,600 message bits in

one image.

6) Steganography and cryptography can be applied in

combination (as showed above). Since the message bits

(i.e. the number of 0’s and 1’s) need to be balanced,

encryption techniques like Vigenere Square of One-

Time-Pad can help evenly spread the message binarily

between 0 and 1.

CLAIR GUTHRIE was born in Fairfax,

Virginia, USA in 1961.

He received a B.S. in Mechanical

engineering from West Virginia

University in 1983 and M.S. in System

Engineering from George Mason

University in 1999. Mr. Guthrie is

pursuing a M.S. in Electrical and

Computer Engineering from George Mason University.

This work was supported by Professor Gaj from George

Mason University Electrical and Computer Engineering

Department.

REFERENCES

[1] Jessica Fridrich, “Steganography in Digital Media

Principles, Algorithms, and Applications,”

Binghamton University , State University of New

York (SUNY), Cambridge University Press, 2010,

Chapter 1, pp. 4-11

[2] Aryfandy Febryan, Tito Saluyo Purboyo and Rady

Erfa Saputra, “Steganogrphy Methods on Text,

Audio, Image and Video: A Survey”, International

Journal of Applied Engineering Research, ISSN

0973-4562 Volume 2, Number 21 (2017),

https://www.ripublication.com/ijaer17/ijaerv12n21_

04.pdf

[3] A. Jabbar, S. Sahib, and M. Zamani , "An Introduction

to Image Steganography Techniques" , Conference

Paper – November 2012 , [Online]. Available:

http://www.reasearchgate.net/publication/258324712

[4] G. J. Simmons. “The Prisoner’s Problem and the

subliminal channel.” In D. Chaum editor, Advances

in Cryptology, CRYPTO ’83, pages 51-67, Santa

Barbra, CA, August 22-24, 1083. New York:

Plenum Press

[5] Gurmeet Kaur and Aarti Kochhar, “A

Steganography Implementation based on LSB &

DCT”, International Journal for Science and

Emerging Technologies with Latest Trends”, ISSN

No. 2277-8136,

https://pdfs.semanticscholar.org/c95d/c820b52ab8d

0dfd0a5f36e99cd40947076f6.pdf

[6] Pooja Shinde and Dr. Tasneem Rehman, “A Survey :

Video Steganography Techniques”, International

Journal of Engineering Research and General

Science, Vol 3, Issue 3, ISSN 2091-2730,

https://www.ijcaonline.org/archives/volume169/nu

mber7/27995-2017914786

[7] N. Meghanathan, and L. Nayak, “Steganalysis

Algorithms for Detecting the Hidden Information in

Image, Audio and Video Cover Media” , in

International Journal of Network Security & Its

Application (IJNSA), Vol.2, No.1, January 2010 ,

https://pdfs.semanticscholar.org/2724/3ae662027ff796

07c556c3127a10e79461b9.pdf

[8] Grayscale, Wikipedia,

https://en.wikipedia.org/wiki/Grayscale

[9] K. Silpa, Dr. S. Aruna Mastani, “Comparison of

Image Quality Metrics”, IJERT, Vol 1 Issue 4, June

2012 ISSN: 2278-0181

[10] Zhou Wany, Alan Bovik, Hamid Sheikh, and Eero

Simoncelli, “Image Quality Assessment: From

Error Visibility to Structural Similarity”, IEEE

Transaction on Image Processing, Vol 13, No. 4,

April 2004,

http://www.cns.nyu.edu/pub/lcv/wang03-

preprint.pdf

[11] “Image Compression: How Math Led to the

JPEG2000 Standard”,

http://www.whydomath.org/node/wavlets/basicjpg.

html

[12] Harpreet Kaur and Jyoti Rani, “A Survey on

different techniques of steganography”, METAC

Web of Conferences 57, ICAET-2016,

https://www.researchgate.net/publication/30297753

2_A_Survey_on_different_techniques_of_steganog

raphy