- IJETR022736
- A More Secure Steganography Method in Spatial Domain.pdf
- 03-Vol8
- A Generalization of the PVD Steganographic Method
- 5_2
- A Survey on Image Steganography using LSB Substitution Technique
- jpeg.pdf
- Steganography
- 631F01_lec9
- Steganography Doc
- v27-5
- ch_01
- Skin Tone Based Secret Data Hiding in Images Using Wavelet Transform
- Steganography
- seminar 459
- Context Based Predictor on Lossless Compression for Android Mobile Devices
- Using Digital Watermarking for Copyright Protection
- BIT 308 Multimedia Question Bank
- c and Mysql Ado Net With Crystals
- A New Revisited Compression Technique Through Innovative Partition Group Binary
- Computer
- Performance Analysis of Image Compression Methods
- IT T71 - MS.doc - QB.doc
- Stego
- Id 2514581462
- ramakanth RESUME
- Potts Ste i Dlt Asche
- 04550696
- Performance Analysis on Wear and Friction Characteristics of Brake Rotor Made of A359-B4c Composites
- A RSA- DWT Based Visual Cryptographic Steganogrphy Technique by Mohit Goel
- IJETAE_0112_29
- IJETAE Certificate 0112 29
- High-Performance JPEG Steganography Using Complementary Embedding Strategy
- 17-spatial
- 1 algo

**Ramakrishna Rao / Indian Journal of Computer Science and Engineering
**

Vol. 1 No. 3 229-239

A Steganography method for JPEG2000 Baseline System

P.Ramakrishna Rao M.Tech.,[CSE], Teaching Associate, Department of Computer Science, Dr.B.R.Ambedkar University,

Etcherla – Srikaulam, 532 410.

Abstract—Hiding capacity is very important for efficient covert communications. For JPEG2000 compressed

images, it is necessary to enlarge the hiding capacity because the available redundancy is very limited. In addition,

the bitstream truncation makes it difficult to hide information. In this paper, a high-capacity steganography scheme

is proposed for the JPEG2000 baseline system, which uses bit-plane encoding procedure twice to solve the problem

due to bitstream truncation. Moreover, embedding points and their intensity are determined in a well deﬁned

quantitative manner via redundancy evaluation to increase hiding capacity. The redundancy is measured by bit,

which is different from conventional methods which adjust the embedding intensity by multiplying a visual masking

factor. High volumetric data is embedded into bit-planes as low as possible to keep message integrality, but at the

cost of an extra bit-plane encoding procedure and slightly changed compression ratio. The proposed method can be

easily integrated into the JPEG2000 image coder, and the produced stego-bitstream can be decoded normally.

Simulation shows that the proposed method is feasible, effective, and secure.

1. INTRODUCTION

MODERN information hiding technology is an important branch of information security. The redundancy

of digital media, as well as the characteristic of human visual system, makes it possible to hide messages.

Information hiding technology used in covert communication is named as steganography. Steganography has

obvious difference with encryption because encryption hides information contents whereas steganography hides

information existence. Three competing aspects, including capacity, security, and robustness, are usually considered

in the designing of information hiding schemes. Security means invisibility and keeping undetectable. Capacity

refers to the maximal secure payload. Robustness relates to the amount of modiﬁcation the stego-object can

withstand before an adversary can destroy the hidden information. Generally speaking, robustness is often

emphasized in applications of digital watermarking rather than steganography. Security and enough hiding capacity

should be needed for desired steganography algorithms. On the one hand, sufficient secret message bits can be

embedded into cover objects, so as to ensure the effectiveness of communications. On the other hand, observable

changes of cover objects ought to be avoided after information embedding, so as to ensure the security of

communications.

The least signiﬁcant bits (LSB) method is widely used to hide data into digital images because of its large

capacity and easy implementation. In this kind of approach, messages are embedded into least signiﬁcant bits of

image pixels, palette indices, or quantized discrete cosine transform (DCT) coefficients. There have been many

steganographic techniques utilizing LSB method, such as EZ-Stego, J-Steg, JPHide-Seek, and Out-Guess.

In JPEG coding system, quantized DCT coefficients are entropy encoded without distortion to get the ﬁnal

compressed bitstream. Secure information hiding can be achieved simply by modiﬁcation on the quantized DCT

coefficients. A DCT domain hiding scheme can be applied in JPEG very conveniently. There have been many kinds

of DCT domain information hiding schemes developed for JPEG standard, such as the above-mentioned J-Steg,

JPHide-Seek, and OutGuess. However, the situation is quite different for JPEG2000. As the latest still image coding

international standard, JPEG2000 is based on discrete wavelet transform (DWT) and embedded block coding and

optimized truncation (EBCOT) algorithms. It offers superior compression performance to JPEG, and puts emphasis

on scalable compressed representations. In JPEG2000 coding system, bitstream is rate-distortion optimizing

truncated after bit-plane encoding. The secret message will be destroyed by the truncating operation if it is

embedded directly into the lowest bit-plane of quantized wavelet coefficients. Although there exist many kinds of

DWT domain hiding schemes, most of them can not be ﬁtted into JPEG2000 directly.

Spread spectrum hiding techniques can be applied in JPEG2000 directly, without consideration on

bitstream truncation. The receiver extracts the hidden information by correlation detection. It is not necessary to

keep all the embedded messages available in correlation detection. However, spread spectrum preprocessing will

decrease hiding capacity signiﬁcantly. Therefore, spread spectrum technology is often used in digital watermarking

applications rather than covert communications.

ISSN : 0976-5166 229

P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering

Vol. 1 No. 3 229-239

For JPEG2000 compressed images, limited redundancy and bitstream truncation makes it difficult to hide

information. After analyzing the challenge of covert communication in JPEG2000 image codec, Su and Kuo

presented a steganography scheme to hide high volumetric data into JPEG2000 bitstream. In order to avoid affection

of bitstream truncation, their method was not designed for the standard baseline system of JPEG2000. It was limited

to the simpliﬁed version of JPEG2000, named as “lazy” mode, in which the entropy coding procedure was

completely bypassed. High-capacity hiding techniques for JPEG2000 standard baseline system should be further

studied. Security is another important property of a desired hiding scheme. As the counterpart of steganography,

steganalysis is the art of detecting steganography. The steganographic method is assumed to be publicly known with

the exception of a secret key. If there exists an algorithm that can guess whether or not a given image contains a

secret message with a success rate better than random guessing, the steganographic system is considered broken.

Excellent work has been done to discover the existence of hidden information. The RS algorithm and the PoVs

steganalysis work well for detecting LSB embedding. Stegdetect is a famous tool that can analyze JPEG images to

detect secret messages hidden by JSteg, JPHide, and Outguess. Farid developed a universal blind detection scheme

that can be applied to any steganographic schemes after proper training on databases of original and cover-images.

In his work, a high-dimensional feature vector was constructed from higher order statistics of wavelet coefficients

and their linear prediction errors. However, it is undesirable to use too many features in terms of classiﬁcation

performance due to the curse of dimensionality.

Compared with spatial or DCT domain steganalysis methods, wavelet domain universal steganalysis

methods ought to be more sensitive to wavelet domain steganography, because the features are extracted from

wavelet domain adopted to verify the security of our hiding scheme.

In this study, a high-capacity steganography scheme is proposed for the commonly used baseline mode of

JPEG2000. This method uses bit-plane encoding procedure twice to solve the problem due to bitstream truncation.

Moreover, the embedding points and their intensity are adjusted image adaptively based on redundancy evaluation

to increase hiding capacity.

2. JPEG2000 BASELINE CODING SYSTEM

The quantized subband is divided into codeblocks that are units of bitplane encoding, with typical

dimensions of 16×16 or 32×32 or 64×64. Bit-plane encoding is operated bit-plane by bit-plane, from high to low, to

produce independent bitstream for each block. Each bit-plane is encoded in a sequence of three fractional bitplane

coding passes. Then, an adaptive arithmetic coding strategy, known as the MQ coder, is employed to encode the

bitstream. This MQ coder is bypassed in the lazy mode of JPEG2000. The EBCOT algorithm produces a ﬁnely

embedded bitstream with many useful truncation points. The bitstream can be truncated at the end of any coding

passes to get desired compression ratio. The truncation point of every codeblock is determined by rate-distortion

optimization.

Source image Wavelet decomposition Quantization

Bit-plane coding Rate-distortion optimization Code stream

Figure1: JPEG2000 baseline encoder

JPEG2000 uses uniform scalar quantizers with enlarged “deadzones.” Truncating the embedded bitstream

associated with any given codeblock has the effect of quantizing the wavelet coefficients in that codeblock more

coarsely. That is to say, there still exists a lossy procedure after entropy encoding. Su and Kuo have pointed out this

problem. It is necessary to take some measures to keep embedded messages available.

ISSN : 0976-5166 230

P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering

Vol. 1 No. 3 229-239

3. STEGANOGRAPHY BASED ON TWICE BIT-PLANE ENCODING

The coding procedure of the proposed method is shown in Figure2.

Source image Wavelet decomposition Quantization

Bit-plane coding Rate-distortion optimization

Determine embedding points and intensity

Message embedding Secondary bit plane encoding Code stream

Figure2: Information hiding based on twice bit-plane encoding

Detailed descriptions of the three additional steps are:

1. There are three sub-steps involved in the determination of embedding points and embedding intensity for a

codeblock as follows:

• The wavelet coefficients greater than a given threshold are chosen as candidate embedding points.

• According to the rate distortion optimization, the lowest bit-plane which keeps unabridged after bitstream

truncation is determined as the lowest embed allowed bitplane of the codeblock.

• The embedding points and embedding intensity are adjusted adaptively on the basis of redundancy

evaluation.

2. Scrambled synchronization information and secret messages are embedded into the selected embedding

points from the lowest embed-allowed bit-plane to higher ones.

3. Secondary bit-plane encoding is operated after information embedding.

After embedding, we organize the bitstream according to the previous result of rate-distortion optimization.

By doing this, messages are embedded into bit-planes that would not be truncated by rate-distortion optimization.

The integrality of the embedded message is ensured at the cost of increased computational complexity and slightly

changed compression ratio. The twice bit-plane encoding procedure is explained to execute the bit-plane encoding

twice, whereas the rest parts, such as wavelet transform, quantization, rate-distortion optimization, bitstream

organization, are executed only once. Information extraction is illustrated in Figure3. First, the lowest bit-plane with

complete information of all its three coding passes can be determined easily in the procedure of entropy decoding.

Then the embedding points and their intensity are determined by the method similar to the encoder. Finally, both

synchronization information and secret messages are extracted.

ISSN : 0976-5166 231

P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering

Vol. 1 No. 3 229-239

Code stream

Image

Figure3: Procedures of message extraction

IV. REDUNDANCY EVALUATION

The redundancy of uniform quantization is evaluated according to the visual masking effect and brightness

sensitivity of human visual system. In this section, wavelet coefficients are processed to do redundancy evaluation,

but not to be encoded. The calculation on self-contrast effect and neighborhood masking effect has been speciﬁed in

the extended system of JPEG2000 for realizing nonuniform quantization. The extended part of JPEG2000 standard

is consulted to select parameter values in the ﬁrst two steps:

1) Self-contrast masking effect is taken in to consideration

y

i

= sign (xֿ

i

) ׀xֿ

i

. Δ

i

׀

α

(1)

where xֿ

i

the quantized wavelet coefficient with the bits is lower than the highest no-zero bit are replaced

by zeros. By clearing the lower bits, we can get identical result in both encoder and decoder. The parameter

Δ

i

is the quantization step of the wavelet coefficient. The parameter assumes a value between 0 and 1. The

result of the first step is y

i

.

2) the neighborhood masking effect is exploited to process the wavelet

Coefficients as the following:

y

i

z

i

= (2)

1 + ( α ∑

k Є neighborhood

׀x

k

׀

β

) / ׀Ø

i

׀

The neighborhood contains wavelet coefﬁcients within a window of N×N, centered at the current position.

The parameter |Ø

i

| is the total number of wavelet coefﬁcients in the neighborhood. The parameter assumes

a value between 0 and 1, together with |Ø

i

|, is used to control the strength of embedding intensity

adjustment due to neighborhood masking. The parameter α is a normalization factor with a constant value

of (10000/2

d-1

)

β

, and d denotes the bit depth of the image component. A small value of β suppresses the

contribution of a few large wavelet coefﬁcients around sharp edges. The parameter β is set to 0.2 in our

experiments. The symbol denotes the neighboring wavelet coefﬁcients greater than or equal to 16, and all

its bits lower than the highest no-zero bit are set to be zeros. The result of the second step is z

i

.

3) a weighting factor about brightness sensitivity is used in the processing. The symbol I

θ

l

denotes the

subband at resolution level lє{0,1,…k} and with orientation θє{LL,LH,HL,HH}.The symbol I

θ

l

(i,j) denotes

the wavelet coefﬁcient located at (i,j) in subband I

θ

l

. The level of discrete wavelet decomposition is k. Level

k is the lowest resolution level, and level 0 is the highest resolution level. Because human eyes are less

sensitive to noise in bright or dark image regions, the local brightness weighting factor Λ(l,i,j) can be

calculated as follows:

Determining embedding points and intensity Message extraction

Parse code stream entropy decoding

Dequantization Wavelet reconstruction

ISSN : 0976-5166 232

P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering

Vol. 1 No. 3 229-239

Λ(l, i, ]) = |

2 -I(l, i, ]): i¡ I(l, i, ]) < 1

I(l, i, ]): tbcr wirc

| (3)

I(l, i, ]) = 1 +

1

128

I

0

k

I 1 +|

ì

2

k-l

| . 1 + |

]

2

k-l

|] (4)

Because the direct current (DC) level of the image signal has been shifted to zero before wavelet

decomposition, the pixel value has a dynamic range of [-128, 127]. The local average brightness is normalized by

dividing 128. Then the result of the third step, ź

i

, is given by

Z

ì

i

=

z

i

A (I,ì,])

(5)

Quantization redundancy is calculated by the following equation:

y

ì

=

x

i

z

i

¡

(6)

Uniform quantization is used in the JPEG2000 baseline system. The redundancy of the wavelet coefﬁcient x`

i

can be

measured by r

i

. In order to reduce the image degradation, we only use the wavelet coefﬁcients with r

i

not less than 2

to carry message bits. The rule of adjustment on embedding points and intensity is as follows:

1) If r

i

< 2, then this candidate embedding point should be removed.

2) If 2

n

≤ r

i

< 2

n+1

, then the embedding capacity of this point is determined to be bits.

V. SYNCHRONIZATION INFORMATION AND SCRAMBLING MEASURE

The secret message must be divided into small fragments before it is embedded into number of codeblocks

of a cover image. Some kind of predeﬁned synchronization information is necessary for accurately extraction of the

hidden message. The synchronization information can be simply structured, as shown in Figure 4.

Figure4: First structure of the Synchronization information

Synchronization information is embedded into every code block before the secret message. The ﬁrst part of

the synchronization information is a 2-bit ﬂag that indicates whether a certain code block contains secret message.

The ﬂag can be set to “11” or “00,” that means “accept” or “reject,” respectively. Only double zeros are to be

embedded into a codeblock when it has too small hiding capacity to hold the synchronization information. The

decoder will be informed by the ﬂag to give up extracting from this code block. The second part of the

synchronization information is a 12-bit fragment that indicates the length of the secret message embedded in this

codeblock.

A more complex structure, depicted in Figure5, can be deﬁned for synchronizing to provide better

performance. After the ﬂag, the follow-up twenty bits indicate where this message fragment localizes in the whole

secret message. The third part of the synchronization information is a 12-bit fragment that indicates the length of the

secret message embedded in this codeblock. For example, if the message fragment for a certain codeblock is from

the n

th

bit to the m

th

bit of the secret message, then the 20-bit fragment should take value of n, and the 12-bit

fragment should take value of (m-n+1). By adding the fragment of localization information, we give error resilience

capability to the steganography scheme. For simpliﬁcation, both the synchronization information bits and the secret

message bits are all called message bits in the following discussion.

ISSN : 0976-5166 233

P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering

Vol. 1 No. 3 229-239

Figure5: Second structure of the synchronization information

A 64-bit secret key is used as a seed to generate a sequence of pseudo random binary numbers, which is

used to scramble the message bits

S

ì

= m

ì

÷ n

ì

: i = 1,2, ……………. . N (7)

where N is the total number of message bits. The symbol m

i

denotes the i

th

message bit, and n

i

the i

th

binary number

of the pseudo random sequence. The operator Ө denotes binary addition. The scrambled message bits, denoted as s

i

,

are to be embedded into selected wavelet coefﬁcients. The same pseudo random sequence, controlled by the secret

key, is generated and used at the decode side to retrieve m

i

simply by an operation of binary addition.

VI. SIMULATION

The source image is Lena, a grayscale image of 512×512, shown in Figure6. Each pixel has eight bits. The

secret message to be embedded is the logo of Civil Aviation University of China, which is shown in Figure6. It is

Figure6: (a) Original image used as cover media (b) the binary logo image used as secret message

a 80×80 binary image with a pixel depth of just one bit. Therefore, there are 6400 bits of secret message in total. In

the following procedure, the source image is compressed at a ratio of 16, and information hiding is conducted

simultaneously.

The source image is decomposed using ﬁve-level DWT. The forth and ﬁfth decomposition levels contain

important low frequency information that can not be modiﬁed to hide secret messages. Only the ﬁrst, second and

third decomposition levels

are used to embed secret messages. The procedures of wavelet decomposition, uniform quantization, bit-plane

encoding and rate distortion optimization have been speciﬁed in JPEG2000 recommendations. We mainly discuss on

the hiding procedure.

1) The lowest embed-allowed bit-plane of each code block is determined. The example is a code block

which locates in the HL subband of the second decomposition level. According to rate-distortion

optimization, the bitstream of this code block should be truncated after the third coding pass of its sixth

bit-plane, counting from the highest nonzero bit-plane. All three coding passes of the sixth bit-plane

will be contained in the ﬁnal bitstream. This bit-plane can keep unabridged even after bitstream

truncation. So the sixth bit-plane should be determined as the lowest embed-allowed bit-plane of this

codeblock. Otherwise, if one or more coding passes of the sixth bit-plane has to be truncated, then the

ﬁfth bit-plane should be chosen as the lowest embed-allowed bit-plane.

2) The wavelet coefficients with magnitudes not less than a given threshold are chosen as candidate

embedding points. Because the low bits will be modiﬁed by information embedding, the threshold

should be the n

th

power of two. This ensures that the embedding points chosen by the receiver are

ISSN : 0976-5166 234

P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering

Vol. 1 No. 3 229-239

consistent with that of the sender. In our experiments, the threshold is set to 16. With four typical

wavelet coefﬁcients as examples, the embedding points are shown in Figure7.

Figure7: candidate embedding points

The wavelet coefficient C can not be chosen to be a embedding point, because it is below the given

threshold. Wavelet coefficients A, B, and D are chosen to be candidate embedding points, whose bits on the

sixth bit-plane are labeled to be embed-allowed. The star pentagon in Figure7 denotes the position into

which allow to embed message bits.

3) The candidate embedding points are adjusted image adaptively based on redundancy evaluation to

increase hiding capacity. The parameters in those equations are set to be: N=5, α =0.7, β=0.2.

Evaluation results for wavelet coefﬁcients A, B, and D are as follows:

r

A

= 1.72 r

B

= 2.59 r

D

= 13.4

as 1.72 < 2, so that wavelet coefﬁcient A can not be used to embed message bits. It should be removed

from the group of candidates, and consequently, the star pentagon marked on it should be erased.

as 2

1

< 2.59 < 2

2

, so that wavelet coefﬁcient B can be used to embed message bits. But its embedding

intensity can not be enlarged any more.

As 2

3

< 13.4 < 2

4

, so that wavelet coefﬁcient D can be used to embed message bits, and extra two bits

can be embedded into wavelet coefﬁcient D. Therefore, the bits on the ﬁfth and forth bit-planes of wavelet

coefﬁcient D are labeled with star pentagons to allow message embedding.

The result of adaptively adjustment is shown in Figure8.

ISSN : 0976-5166 235

P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering

Vol. 1 No. 3 229-239

Figure8: Finally adjusted embedding points and their intensity

4) In the fourth step, we embed message bits into the selected wavelet coefﬁcients and ﬁnish encoding the

stegoimage. There are totally 63 code blocks in the ﬁrst, second, and third decomposition levels of the test

image. The embedding payload of each code block can be calculated to show roughly direct proportion to its

hiding capacity. After payload assignment, synchronization information is added and message scrambling is

conducted. Then, the code block is scanned bit-plane by bit-plane, from low to high, beginning with the lowest

embed-allowed bit-plane. Once it comes across an embed-allowed position, one message bit is embedded, until

all the message bits are embedded. After message embedding, secondary bit-plane encoding is operated on the

modiﬁed code blocks, and, ﬁnally, the ﬁnal bitstream is organized.

The bitstream can be decoded using a standard JPEG2000 decoder. The reconstructed image is shown in

Figure9, which can not be distinguished from the original one shown in Figure6. Information extraction

Figure9: (a) Reconstructed Lena image (b) the retrieved logo image

is just the reverse operation of information embedding. Experiment shows that information hiding has caused slight

change on PSNR (Peak Signal to Noise Ratio) and the actual compression ratio. Because information hiding

ISSN : 0976-5166 236

P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering

Vol. 1 No. 3 229-239

techniques are often based on the visual masking effect of human visual system, it is not suitable to use PSNR to

evaluate the stegoimage degradation. The most reasonable method for evaluating subjective quality is by

observation, which is also in common use. We have many people vote for the visible degradation of the stego-image

to improve hiding capacity measurement. Beginning with a small quantity, the hidden message is increased by 100

bits each time. The capacity is obtained when the stego-image degradation becomes visible. The slight change of

compression ratio is due to the bitstream of the secondary bit-plane encoding is different from the initial one. This

small change, less than one percent in our experiments, is usually acceptable.

In order to test and measure the effectiveness on hiding capacity enlargement, we simply bypass the

redundancy evaluation for comparison. Two methods are tested in the experiments.

• Method 1: With redundancy evaluation.

• Method 2: Without redundancy evaluation.

The two methods should be tested with different cover images and different

Figure10: Two images with different texture activity: (a) Crown (b) Baboon

Table 1: Hiding capacity of three test images

Crown Lena Baboon

Method 2 3000 6500 11000

Method 1 4500 14000 19500

compression ratios. Together with Lena, two more images shown in Figure10 are used. The three images have same

size, same pixel depth, and different texture activity. Both the two methods are tested on these images, with a

compression ratio of 0.8 bits per pixel. The results are listed in Table I. The hiding capacity of Lena at different

compression ratios is shown in Figure11. It can be seen that the proposed method is effective, especially for those

images with uneven brightness and diverse texture activity.

ISSN : 0976-5166 237

P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering

Vol. 1 No. 3 229-239

Figure11: Hiding capacity of different compression ratios

The proposed method hides information by modifying wavelet coefﬁcients. Accordingly, a wavelet domain

steganalysis method should be the best choice for the security veriﬁcation. The universal steganalysis method

described in, which extracts features from wavelet domain, is used to do the security veriﬁcation. The stego image

estimation method used has been detailed described in the author’s earlier work reported. In the following

experiment, samples are 1100 never compressed gray-scale digital pictures with resolution of 640480. Four

examples of these pictures are shown in Figure12. We randomly choose 800 original images and the corresponding

800 stego ones for calculating the projection vector of the FLD (Fisher Linear Discriminator) classiﬁer. The

remaining 300 original images and the

Figure 12: Four different images

corresponding 300 stego ones are used for testing purpose. Figure13 shows ROC (Receiver Operating

Characteristic) curves tested on different payloads: 15000 bits, 25000 bits, 35000 bits, 60000 bits, and 80000 bits.

According to the experimental data, the detector is in vain when the message length is 15000 or 25000. And when

we increase the payload to 35000 bits,

ISSN : 0976-5166 238

P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering

Vol. 1 No. 3 229-239

Figure13: ROC carves tested on different payloads

there is still no obvious detection effect. The detector does work only if the message length greatly exceeds the

hiding capacity. In general case, the proposed steganography scheme can be considered undetectable in the situation

of lower payloads than hiding capacity.

VII. CONCLUSION

In this study, a high-capacity steganographic scheme has been proposed for the JPEG2000 baseline system.

The contributions of this work are mainly focused on dealing with two problems: bitstream truncation and

redundancy measurement. The steganographic encoding procedure is explained to execute bit-plane encoding twice,

whereas the rest parts, such as wavelet transform, quantization, rate-distortion optimization, bitstream organization,

still are executed only once. Therefore, the computational complexity hasn’t been increased too much.

REFERENCES

[1] C. C. Chang,Y. C. Chou, and T.D.Kieu, “High capacity data hiding for grayscale images,” in Proc. 1st Int. Conf. Ubiquitou

Information Management and Communication, Seoul, Korea, Feb. 2007, pp. 139–148.

[2] S.A.M.Gilani, I.Kostopoulos, andA.N. Skodras, “Color image-adaptive watermarking,” in Proc. 14th Int. Conf. Digital

Signal Processing, 2002, vol. 2, pp. 721- 724.

[3] C. K. Chan and L. M. Cheng, “Hiding data in images by simple LSB substitution,” Pattern Recognit., vol. 37, no. 3, pp.

469–474, 2004.

[4] J. Fridrich and M. Goljan, “Practical steganalysis of digital imagesState of the art,” in Proc. SPIE, 2002, vol. 4675, pp. 1–13.

[5] JPEG2000 Part 1: Final Committee Draft Version 1.0, ISO/IEC. FCD 15444-1, 2000.

[6] JPEG2000 Part 2: Final Committee Draft, ISO/IEC FCD 15444-2,2000.

[7] A. Westfeld and A. Pﬁtzmann, “Attacks on steganographic systems,”in Lecture Notes in Computer Science. Berlin,

Germany: Springer- Verlag, 2000, vol. 1768, pp. 61–75.

[8] H. Farid, “Detecting hidden messages using higher-order statistics models,” in Proc. EEE Int. Conf. Image Processing, New

York, 2002, pp. 905–908.

[9] T. Holotyak, J. Fridrich, and S. Voloshynovskiy, “Blind statistical steganalysis of additive steganography using wavelet

higher order statistics,” presented at the 9th IFIP TC-6 TC-11 Conf. Communications and Multimedia Security, 2005.

ISSN : 0976-5166 239

an adaptive arithmetic coding strategy. is employed to encode the bitstream. In this study. because the features are extracted from wavelet domain adopted to verify the security of our hiding scheme. ISSN : 0976-5166 230 . The truncation point of every codeblock is determined by rate-distortion optimization. JPHide. from high to low. The bitstream can be truncated at the end of any coding passes to get desired compression ratio. 3 229-239 For JPEG2000 compressed images. their method was not designed for the standard baseline system of JPEG2000. wavelet domain universal steganalysis methods ought to be more sensitive to wavelet domain steganography. limited redundancy and bitstream truncation makes it difficult to hide information. Farid developed a universal blind detection scheme that can be applied to any steganographic schemes after proper training on databases of original and cover-images. This MQ coder is bypassed in the lazy mode of JPEG2000. the embedding points and their intensity are adjusted image adaptively based on redundancy evaluation to increase hiding capacity. However. After analyzing the challenge of covert communication in JPEG2000 image codec. Stegdetect is a famous tool that can analyze JPEG images to detect secret messages hidden by JSteg. JPEG2000 BASELINE CODING SYSTEM The quantized subband is divided into codeblocks that are units of bitplane encoding. Excellent work has been done to discover the existence of hidden information.P. there still exists a lossy procedure after entropy encoding. Moreover. Ramakrishna Rao / Indian Journal of Computer Science and Engineering Vol. 2. Bit-plane encoding is operated bit-plane by bit-plane.” Truncating the embedded bitstream associated with any given codeblock has the effect of quantizing the wavelet coefficients in that codeblock more coarsely. steganalysis is the art of detecting steganography. In order to avoid affection of bitstream truncation. a high-capacity steganography scheme is proposed for the commonly used baseline mode of JPEG2000. High-capacity hiding techniques for JPEG2000 standard baseline system should be further studied. named as “lazy” mode. with typical dimensions of 16×16 or 32×32 or 64×64. If there exists an algorithm that can guess whether or not a given image contains a secret message with a success rate better than random guessing. Source image Wavelet decomposition Quantization Bit-plane coding Rate-distortion optimization Code stream Figure1: JPEG2000 baseline encoder JPEG2000 uses uniform scalar quantizers with enlarged “deadzones. in which the entropy coding procedure was completely bypassed. Then. Security is another important property of a desired hiding scheme. That is to say. The RS algorithm and the PoVs steganalysis work well for detecting LSB embedding. The steganographic method is assumed to be publicly known with the exception of a secret key. Su and Kuo have pointed out this problem. and Outguess. As the counterpart of steganography. 1 No. to produce independent bitstream for each block. It was limited to the simpliﬁed version of JPEG2000. a high-dimensional feature vector was constructed from higher order statistics of wavelet coefficients and their linear prediction errors. This method uses bit-plane encoding procedure twice to solve the problem due to bitstream truncation. The EBCOT algorithm produces a ﬁnely embedded bitstream with many useful truncation points. it is undesirable to use too many features in terms of classiﬁcation performance due to the curse of dimensionality. It is necessary to take some measures to keep embedded messages available. Su and Kuo presented a steganography scheme to hide high volumetric data into JPEG2000 bitstream. Compared with spatial or DCT domain steganalysis methods. the steganographic system is considered broken. Each bit-plane is encoded in a sequence of three fractional bitplane coding passes. In his work. known as the MQ coder.

• According to the rate distortion optimization. Information extraction is illustrated in Figure3. Ramakrishna Rao / Indian Journal of Computer Science and Engineering Vol. 2.P. Source image Wavelet decomposition Quantization Bit-plane coding Rate-distortion optimization Determine embedding points and intensity Message embedding Secondary bit plane encoding Code stream Figure2: Information hiding based on twice bit-plane encoding Detailed descriptions of the three additional steps are: 1. bitstream organization. By doing this. we organize the bitstream according to the previous result of rate-distortion optimization. rate-distortion optimization. 1 No. First. quantization. ISSN : 0976-5166 231 . the lowest bit-plane with complete information of all its three coding passes can be determined easily in the procedure of entropy decoding. the lowest bit-plane which keeps unabridged after bitstream truncation is determined as the lowest embed allowed bitplane of the codeblock. The twice bit-plane encoding procedure is explained to execute the bit-plane encoding twice. Finally. STEGANOGRAPHY BASED ON TWICE BIT-PLANE ENCODING The coding procedure of the proposed method is shown in Figure2. After embedding. Scrambled synchronization information and secret messages are embedded into the selected embedding points from the lowest embed-allowed bit-plane to higher ones. such as wavelet transform. The integrality of the embedded message is ensured at the cost of increased computational complexity and slightly changed compression ratio. There are three sub-steps involved in the determination of embedding points and embedding intensity for a codeblock as follows: • The wavelet coefficients greater than a given threshold are chosen as candidate embedding points. 3 229-239 3. • The embedding points and embedding intensity are adjusted adaptively on the basis of redundancy evaluation. messages are embedded into bit-planes that would not be truncated by rate-distortion optimization. Secondary bit-plane encoding is operated after information embedding. whereas the rest parts. are executed only once. 3. both synchronization information and secret messages are extracted. Then the embedding points and their intensity are determined by the method similar to the encoder.

The parameter assumes a value between 0 and 1. 3) a weighting factor about brightness sensitivity is used in the processing. The result of the first step is yi. The parameter β is set to 0. and level 0 is the highest resolution level.HH}.P. Because human eyes are less sensitive to noise in bright or dark image regions. wavelet coefficients are processed to do redundancy evaluation.2 in our experiments. The parameter α is a normalization factor with a constant value of (10000/2d-1)β. REDUNDANCY EVALUATION The redundancy of uniform quantization is evaluated according to the visual masking effect and brightness sensitivity of human visual system. 1 No. The result of the second step is zi. centered at the current position.i. The calculation on self-contrast effect and neighborhood masking effect has been speciﬁed in the extended system of JPEG2000 for realizing nonuniform quantization.1. 3 229-239 Code stream Parse code stream entropy decoding Determining embedding points and intensity Message extraction Dequantization Wavelet reconstruction Image Figure3: Procedures of message extraction IV. Ramakrishna Rao / Indian Journal of Computer Science and Engineering Vol. The level of discrete wavelet decomposition is k. The symbol denotes the neighboring wavelet coefﬁcients greater than or equal to 16. and d denotes the bit depth of the image component.j) in subband Iθl. The parameter |Øi| is the total number of wavelet coefﬁcients in the neighborhood.HL. Δi ׀α (1) where xֿi the quantized wavelet coefficient with the bits is lower than the highest no-zero bit are replaced by zeros. A small value of β suppresses the contribution of a few large wavelet coefﬁcients around sharp edges. and all its bits lower than the highest no-zero bit are set to be zeros. The parameter Δi is the quantization step of the wavelet coefficient. 2) the neighborhood masking effect is exploited to process the wavelet Coefficients as the following: yi (2) zi = 1 + ( α ∑k Є neighborhood ׀xk׀β) / ׀Øi׀ The neighborhood contains wavelet coefﬁcients within a window of N×N. we can get identical result in both encoder and decoder.The symbol Iθl(i. is used to control the strength of embedding intensity adjustment due to neighborhood masking. In this section. together with |Øi|. Level k is the lowest resolution level.LH.j) denotes the wavelet coefﬁcient located at (i. the local brightness weighting factor Λ(l.…k} and with orientation θє{LL. The extended part of JPEG2000 standard is consulted to select parameter values in the ﬁrst two steps: 1) Self-contrast masking effect is taken in to consideration yi = sign (xֿi) ׀xֿi . The parameter assumes a value between 0 and 1. but not to be encoded. The symbol Iθl denotes the subband at resolution level lє{0.j) can be calculated as follows: ISSN : 0976-5166 232 . By clearing the lower bits.

The local average brightness is normalized by dividing 128. In order to reduce the image degradation. . By adding the fragment of localization information. can be deﬁned for synchronizing to provide better performance. The rule of adjustment on embedding points and intensity is as follows: 1) If ri< 2. 3 229-239 Λ . The ﬁrst part of the synchronization information is a 2-bit ﬂag that indicates whether a certain code block contains secret message. After the ﬂag. then the embedding capacity of this point is determined to be bits. For simpliﬁcation. depicted in Figure5. 2 . Figure4: First structure of the Synchronization information Synchronization information is embedded into every code block before the secret message. 127]. For example. źi . The third part of the synchronization information is a 12-bit fragment that indicates the length of the secret message embedded in this codeblock. . we only use the wavelet coefﬁcients with ri not less than 2 to carry message bits. 1 (3) 1 . both the synchronization information bits and the secret message bits are all called message bits in the following discussion. if the message fragment for a certain codeblock is from the nth bit to the mth bit of the secret message. the pixel value has a dynamic range of [-128. (5) (6) Quantization redundancy is calculated by the following equation: Uniform quantization is used in the JPEG2000 baseline system. V. 1 No. 1 . and the 12-bit fragment should take value of (m-n+1). ISSN : 0976-5166 233 . Ramakrishna Rao / Indian Journal of Computer Science and Engineering Vol. The decoder will be informed by the ﬂag to give up extracting from this code block. .. as shown in Figure 4. then the 20-bit fragment should take value of n. The second part of the synchronization information is a 12-bit fragment that indicates the length of the secret message embedded in this codeblock. then this candidate embedding point should be removed. Some kind of predeﬁned synchronization information is necessary for accurately extraction of the hidden message. . : . 1 (4) Because the direct current (DC) level of the image signal has been shifted to zero before wavelet decomposition. is given by . the follow-up twenty bits indicate where this message fragment localizes in the whole secret message. Then the result of the third step.” that means “accept” or “reject. SYNCHRONIZATION INFORMATION AND SCRAMBLING MEASURE The secret message must be divided into small fragments before it is embedded into number of codeblocks of a cover image. The redundancy of the wavelet coefﬁcient x`i can be measured by ri.P. . The ﬂag can be set to “11” or “00. . The synchronization information can be simply structured. A more complex structure. .” respectively. we give error resilience capability to the steganography scheme. 2) If 2n ≤ ri< 2n+1. Only double zeros are to be embedded into a codeblock when it has too small hiding capacity to hold the synchronization information.

is generated and used at the decode side to retrieve mi simply by an operation of binary addition. the bitstream of this code block should be truncated after the third coding pass of its sixth bit-plane. denoted as si. So the sixth bit-plane should be determined as the lowest embed-allowed bit-plane of this codeblock. All three coding passes of the sixth bit-plane will be contained in the ﬁnal bitstream. The procedures of wavelet decomposition. (7) where N is the total number of message bits. 1) The lowest embed-allowed bit-plane of each code block is determined. It is Figure6: (a) Original image used as cover media (b) the binary logo image used as secret message a 80×80 binary image with a pixel depth of just one bit. 3 229-239 Figure5: Second structure of the synchronization information A 64-bit secret key is used as a seed to generate a sequence of pseudo random binary numbers. shown in Figure6. controlled by the secret key. bit-plane encoding and rate distortion optimization have been speciﬁed in JPEG2000 recommendations. We mainly discuss on the hiding procedure. the source image is compressed at a ratio of 16. which is used to scramble the message bits . According to rate-distortion optimization. The forth and ﬁfth decomposition levels contain important low frequency information that can not be modiﬁed to hide secret messages. uniform quantization. The secret message to be embedded is the logo of Civil Aviation University of China. there are 6400 bits of secret message in total. The operator Ө denotes binary addition. VI. This ensures that the embedding points chosen by the receiver are ISSN : 0976-5166 234 . Each pixel has eight bits.2. Only the ﬁrst. The symbol mi denotes the ith message bit. 1 No. … … … … … . The example is a code block which locates in the HL subband of the second decomposition level. In the following procedure. the threshold should be the nth power of two. The source image is decomposed using ﬁve-level DWT. counting from the highest nonzero bit-plane. The same pseudo random sequence. . 2) The wavelet coefficients with magnitudes not less than a given threshold are chosen as candidate embedding points. which is shown in Figure6.P. Ramakrishna Rao / Indian Journal of Computer Science and Engineering Vol. and ni the ith binary number of the pseudo random sequence. 1. are to be embedded into selected wavelet coefﬁcients. Because the low bits will be modiﬁed by information embedding. Otherwise. if one or more coding passes of the sixth bit-plane has to be truncated. The scrambled message bits. and information hiding is conducted simultaneously. SIMULATION The source image is Lena. then the ﬁfth bit-plane should be chosen as the lowest embed-allowed bit-plane. Therefore. second and third decomposition levels are used to embed secret messages. This bit-plane can keep unabridged even after bitstream truncation. a grayscale image of 512×512.

so that wavelet coefﬁcient A can not be used to embed message bits. With four typical wavelet coefﬁcients as examples. α =0. whose bits on the sixth bit-plane are labeled to be embed-allowed. so that wavelet coefﬁcient B can be used to embed message bits. It should be removed from the group of candidates.4 rA = 1. In our experiments. Ramakrishna Rao / Indian Journal of Computer Science and Engineering Vol. as 21 < 2.72 as 1.72 < 2. β=0. 3) The candidate embedding points are adjusted image adaptively based on redundancy evaluation to increase hiding capacity. and D are as follows: rB = 2. the star pentagon marked on it should be erased.P. and extra two bits can be embedded into wavelet coefﬁcient D. B. 1 No. Wavelet coefficients A.59 < 22. B. Figure7: candidate embedding points The wavelet coefficient C can not be chosen to be a embedding point. the threshold is set to 16. Evaluation results for wavelet coefﬁcients A. because it is below the given threshold. and D are chosen to be candidate embedding points. The parameters in those equations are set to be: N=5. so that wavelet coefﬁcient D can be used to embed message bits.59 rD = 13. the bits on the ﬁfth and forth bit-planes of wavelet coefﬁcient D are labeled with star pentagons to allow message embedding. Therefore. the embedding points are shown in Figure7. 3 229-239 consistent with that of the sender. ISSN : 0976-5166 235 .2. But its embedding intensity can not be enlarged any more. and consequently.4 < 24 . The star pentagon in Figure7 denotes the position into which allow to embed message bits.7. The result of adaptively adjustment is shown in Figure8. As 23 < 13.

3 229-239 Figure8: Finally adjusted embedding points and their intensity 4) In the fourth step. beginning with the lowest embed-allowed bit-plane. ﬁnally. from low to high. Experiment shows that information hiding has caused slight change on PSNR (Peak Signal to Noise Ratio) and the actual compression ratio. and third decomposition levels of the test image.P. Because information hiding ISSN : 0976-5166 236 . The embedding payload of each code block can be calculated to show roughly direct proportion to its hiding capacity. synchronization information is added and message scrambling is conducted. Then. one message bit is embedded. 1 No. After payload assignment. There are totally 63 code blocks in the ﬁrst. The reconstructed image is shown in Figure9. Ramakrishna Rao / Indian Journal of Computer Science and Engineering Vol. After message embedding. the code block is scanned bit-plane by bit-plane. we embed message bits into the selected wavelet coefﬁcients and ﬁnish encoding the stegoimage. the ﬁnal bitstream is organized. The bitstream can be decoded using a standard JPEG2000 decoder. Once it comes across an embed-allowed position. which can not be distinguished from the original one shown in Figure6. secondary bit-plane encoding is operated on the modiﬁed code blocks. second. until all the message bits are embedded. Information extraction Figure9: (a) Reconstructed Lena image (b) the retrieved logo image is just the reverse operation of information embedding. and.

• Method 2: Without redundancy evaluation. ISSN : 0976-5166 237 . less than one percent in our experiments. In order to test and measure the effectiveness on hiding capacity enlargement. Beginning with a small quantity. which is also in common use. The two methods should be tested with different cover images and different Figure10: Two images with different texture activity: (a) Crown (b) Baboon Table 1: Hiding capacity of three test images Crown Method 2 Method 1 3000 4500 6500 14000 Lena 11000 19500 Baboon compression ratios.8 bits per pixel. Together with Lena. we simply bypass the redundancy evaluation for comparison. 3 229-239 techniques are often based on the visual masking effect of human visual system. same pixel depth. The results are listed in Table I. The most reasonable method for evaluating subjective quality is by observation. Two methods are tested in the experiments.P. it is not suitable to use PSNR to evaluate the stegoimage degradation. We have many people vote for the visible degradation of the stego-image to improve hiding capacity measurement. It can be seen that the proposed method is effective. with a compression ratio of 0. The three images have same size. Both the two methods are tested on these images. especially for those images with uneven brightness and diverse texture activity. This small change. is usually acceptable. The capacity is obtained when the stego-image degradation becomes visible. • Method 1: With redundancy evaluation. and different texture activity. 1 No. The slight change of compression ratio is due to the bitstream of the secondary bit-plane encoding is different from the initial one. two more images shown in Figure10 are used. Ramakrishna Rao / Indian Journal of Computer Science and Engineering Vol. The hiding capacity of Lena at different compression ratios is shown in Figure11. the hidden message is increased by 100 bits each time.

3 229-239 Figure11: Hiding capacity of different compression ratios The proposed method hides information by modifying wavelet coefﬁcients. is used to do the security veriﬁcation. and 80000 bits. ISSN : 0976-5166 238 .P. Four examples of these pictures are shown in Figure12. Accordingly. The universal steganalysis method described in. the detector is in vain when the message length is 15000 or 25000. Ramakrishna Rao / Indian Journal of Computer Science and Engineering Vol. Figure13 shows ROC (Receiver Operating Characteristic) curves tested on different payloads: 15000 bits. a wavelet domain steganalysis method should be the best choice for the security veriﬁcation. In the following experiment. And when we increase the payload to 35000 bits. 35000 bits. samples are 1100 never compressed gray-scale digital pictures with resolution of 640480. 25000 bits. The remaining 300 original images and the Figure 12: Four different images corresponding 300 stego ones are used for testing purpose. The stego image estimation method used has been detailed described in the author’s earlier work reported. 1 No. According to the experimental data. We randomly choose 800 original images and the corresponding 800 stego ones for calculating the projection vector of the FLD (Fisher Linear Discriminator) classiﬁer. which extracts features from wavelet domain. 60000 bits.

2002.Kieu. whereas the rest parts. [7] A. Skodras. 1 No. Communications and Multimedia Security. 2002. VII. a high-capacity steganographic scheme has been proposed for the JPEG2000 baseline system. Conf. “Practical steganalysis of digital imagesState of the art. rate-distortion optimization. [6] JPEG2000 Part 2: Final Committee Draft. ISO/IEC FCD 15444-2. Farid. Feb.2000. Berlin. Westfeld and A. EEE Int. [3] C. 3 229-239 Figure13: ROC carves tested on different payloads there is still no obvious detection effect. andA. FCD 15444-1. vol. C. 2. Digital Signal Processing. M. New York. “Attacks on steganographic systems. pp. ISO/IEC. Goljan.Kostopoulos. 1st Int. In general case. [8] H. 2005. The contributions of this work are mainly focused on dealing with two problems: bitstream truncation and redundancy measurement.Y.”in Lecture Notes in Computer Science. quantization. “Hiding data in images by simple LSB substitution. pp. “Blind statistical steganalysis of additive steganography using wavelet higher order statistics. C. Germany: Springer. K. “High capacity data hiding for grayscale images. 139–148.” in Proc. The steganographic encoding procedure is explained to execute bit-plane encoding twice. 4675. 1–13. 2004. no.” in Proc. [4] J. Ramakrishna Rao / Indian Journal of Computer Science and Engineering Vol. Conf. 2000. Therefore. 37.M. pp. Cheng. 2007. Image Processing. I. vol. Fridrich. The detector does work only if the message length greatly exceeds the hiding capacity. [5] JPEG2000 Part 1: Final Committee Draft Version 1. Pﬁtzmann.D. 905–908. such as wavelet transform. Fridrich and M. “Detecting hidden messages using higher-order statistics models. 721.” presented at the 9th IFIP TC-6 TC-11 Conf. the proposed steganography scheme can be considered undetectable in the situation of lower payloads than hiding capacity.724. pp. pp.. Seoul. and T. Holotyak. vol. Ubiquitou Information Management and Communication.” in Proc. Chou. 3. pp. CONCLUSION In this study. Chan and L. Voloshynovskiy.0. the computational complexity hasn’t been increased too much. 469–474.A. “Color image-adaptive watermarking. bitstream organization. vol. 2002. still are executed only once.P. REFERENCES [1] C.Verlag. [9] T. 1768. Korea.” in Proc.N.” Pattern Recognit. 61–75. 14th Int. and S. Conf. [2] S. Chang. ISSN : 0976-5166 239 .Gilani. SPIE. J. 2000.

- IJETR022736Uploaded byerpublication
- A More Secure Steganography Method in Spatial Domain.pdfUploaded byParamesh Waran
- 03-Vol8Uploaded byResearch Cell: An International Journal of Engineering Sciences
- A Generalization of the PVD Steganographic MethodUploaded byijcsis
- 5_2Uploaded byBleu Oiseau
- A Survey on Image Steganography using LSB Substitution TechniqueUploaded byAnonymous kw8Yrp0R5r
- jpeg.pdfUploaded byphantom_nc91
- SteganographyUploaded bymadhavi461
- 631F01_lec9Uploaded byramulu_492
- Steganography DocUploaded byDinesh Reddy
- v27-5Uploaded bySandeep Das
- ch_01Uploaded byjunaid aslam
- Skin Tone Based Secret Data Hiding in Images Using Wavelet TransformUploaded byATS
- SteganographyUploaded bysumitjain1313
- seminar 459Uploaded bypriyanka07422
- Context Based Predictor on Lossless Compression for Android Mobile DevicesUploaded byInternational Journal for Scientific Research and Development - IJSRD
- Using Digital Watermarking for Copyright ProtectionUploaded byMulkan Cielers Laruku
- BIT 308 Multimedia Question BankUploaded byPal Kumar
- c and Mysql Ado Net With CrystalsUploaded byPhiokham Suriya
- A New Revisited Compression Technique Through Innovative Partition Group BinaryUploaded byIAEME Publication
- ComputerUploaded byTobey Milia
- Performance Analysis of Image Compression MethodsUploaded bySabbir Hossain
- IT T71 - MS.doc - QB.docUploaded byJayanthiSakthi
- StegoUploaded byShrinidhi Gowda
- Id 2514581462Uploaded byAnonymous 7VPPkWS8O
- ramakanth RESUMEUploaded byraamahi
- Potts Ste i Dlt AscheUploaded byistarhir
- 04550696Uploaded byjani442
- Performance Analysis on Wear and Friction Characteristics of Brake Rotor Made of A359-B4c CompositesUploaded byantonytechno
- A RSA- DWT Based Visual Cryptographic Steganogrphy Technique by Mohit GoelUploaded byMohit Goel