Analysis of LSB Algorithm Modification with Bit Inverse and

Insertion based on Length of Message

Fahrul Ikhsan Lubis

, Saib Suwilo

and Poltak Sihombing

Magister Informatics, Universitas Sumatera Utara, Medan, 20155, Indonesia,

Department of Informatic, Universitas Sumatera Utara, 20155, Indonesia

Keywords: Steganography, LSB Algorithm Modification, Bit Inverse, Length of Message, LCG Algorithm, Text

Message, Image Massage.

Abstract: Steganographic demolition using the LSB Algorithm is increasingly being carried out by researchers for

modifications as needed. Previously, the LSB algorithm has been modified by doing a bit inverse where the

message is inserted into the container image and then modified by looking at the 2nd and 3rd LSB bits. If the

two bits are different then the inverse of the entered message bit (bit 1 LSB) is performed. However, if the

two values are the same then the entered message will not be inverted. This modification is considered good

but still lacks in the pixel position of the image that is inserted with the message (modification is done by

inserting sequential images in an image). To cover this weakness, in this study a modification of the LSB with

insertion based on the length of the message was carried out. Message insertion was carried out on random

pixels of an image using the LCG algorithm and the number of message bits that were inserted varies

according to the message length. Moreover, the modifications made should produce a better stegoimage than

before, as evidenced by calculating the PSNR of a stegoimage. From the modification test results, it was found

better results than before both in text message archiving (66.29 dB > 61.8 dB) or in image messages (54.20

dB > 50.01 dB).

1 INTRODUCTION

The insertion of messages into images using

steganography techniques has experienced rapid

development. The researchers competed to make their

own modifications according to their knowledge and

experience of the insertion. This occurs because the

basic theory of message insertion is considered to

have been overused and tried to modify the method.

Previously, the algorithm (LSB) has been

modified by inverse bit where the message will be

inserted into the container image first. Furthermore,

the 2nd and 3rd bit LSB is scanned. If the two bits

have different binaries then the inserted message

binary will be converted to the inverse binary that was

inserted. If the 2nd and 3rd bit LSB are the same, then

the inserted message does not change (Bharwaj and

Sharma, 2016).

Previous research has produced a good message

insertion, but the modification only inserts one

message bit in each pixel and the insertion position is

done on consecutive pixels in an image. This

modification has gaps to be recognized because if a

more detailed check is carried out, the reduction in

taste quality in certain parts in sequence will be

detected. To cover the weakness of the previous

researchers, the second modification inserts messages

on random pixels in an image and the number of

message bits inserted is adjusted to the message bit

length. If the length of the message in modulo 2 is 0,

then 2 message bits will be inserted at each image

layer at positions 1 and 2 LSB. Meanwhile, if the

result is 1, the insertion of 1 message bit on each

image layer at position 1 LSB is carried out. Thus the

insertion will be difficult to detect because the

insertion of random messages using steganographic

techniques will increase the security of the message

(Sitompul et al, 2018) and the number of bits inserted

can change according to the message length.

Steganoimage of the two LSB modifications will

be tested by calculating the Peak Signal to Noise

Ratio (PSNR), Mean Squared Error (MSE), and the

time required to insert the message.

522

Lubis, F., Suwilo, S. and Sihombing, P.

Analysis of LSB Algorithm Modiﬁcation with Bit Inverse and Insertion based on Length of Message.

DOI: 10.5220/0010333500003051

In Proceedings of the International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies (CESIT 2020), pages 522-529

ISBN: 978-989-758-501-2

2 RELATED STUDIES

Steganography was a process to embed data to cover

object. The cover object might be image, text, audio,

or video. Cover object were a container to hidden

data or secret messages and they were the main

material of steganographic systems, where some of

their characteristics were altered or manipulated to

conceal a confidential message. However, these

modifications or manipulation that occurred during

the process of concealing should rest unnoticeable

for anyone not participating in the process of

communication. The ability of files used as a cover

for embedding confidential data depends on the

availability of redundant or unimportant areas inside

these files, so the cover file size might be larger than

the message size to include (Shtayt et al, 2020).

Image steganography was the most widely used,

compared with the other types of steganography. This

popularity was because the images have a large amount

of redundant information. Image steganographic

techniques were evaluated by three principles: (i)

Capacity, The amount of information was hidden

inside the cover file; (ii) Imperceptibility, The

invisibility of the hidden data was in the cover file

without destroying image quality; (iii)Security, How

could a stego file resist the different steganalysis

detection attacks. (Al-Aidroos and Bahamish, 2019)

The terms in steganography were: (-) Embedded

message or Hidden text: secret message to be

inserted; (-). Cover-object: the object of the message

insertion (embedded message); (-) Stegoimage or

Stego-object: objects that had been inserted secret

messages (embedded message); (-) Embedding or

Encoding: the process of inserting a message into the

Cover object; (-) Extraction or Decoding: Stegoimage

extraction process to issue an original message

(embedded message) (Gunawan and Sumarno, 2018).

Steganography techniques were divided into

Spatial (time) domain and Transform (frequency)

domain technique: Spatial Domain Technique, uses

the pixel values of images directly for encoding the

secret message. This class of technique was Least

Significant Bit (LSB) replacement technique in which

firstly binary representation of the image’s pixel

value was calculated then bits were used to hide the

secret messages. Initially, for a 24-bit image, each of

the red, green and blue colour components of bit

could be used, as each was represented by a byte. In

other words, one could store 3 bits in each pixel. This

technique hide the secret message bits into the 3 LSB

bits of the cover image.

Transform Domain Technique, These transform

domain techniques mainly included Discrete Cosine

Transform (DCT), Discrete Wavelet Transform

(DWT) and Discrete Fourier Transform (DFT). This

technique used DCT and Blowfish algorithm in which

the LSB of each DC coefficient replace with each bit

of secret message. This algorithm was proposed to

increase the security of hidden message. (Singh et al,

2018).

2.1 Least Significant Bit (LSB)

Algorithm

LSB algorithm was modifying the last bits in the

cover object in every byte of color in a pixel with

replacing each last bit with the secret bits of

information (Sari and Siahaan, 2018). In the using of

LSB algorithm, firstly the message and the cover

image must be converted to binary, after it completed,

one bit message would be added to each layer in the

cover image pixel. The procedure was as follows:

Suppose the message binary was 010 and the

binary of one pixel image in each layer was:

01010111 01010011 01010111

If the message and binary image were known,

subtitute the message at the end of the binary or the

LSB binary image as follows:

01010110 01010011 01010110

After the pixel binary completed it would be

converted into a matrix and made into an image again.

2.2 Image

Image was a photo or two-dimensional dwasplay that

could describe the vwasualization of objects. The

image could be grouped into print or digital form.

Digital images could be converted into an array of

numbers, while printed images must first be

converted to digital if they want to be processed

Digital image was a collection of thousands of

very small dots and each of these dots has a certain

color. The small boxes were called pixels, the number

of pixels in an image could also determine the size of

the image. Resolution was the number of pixels per

certain. If the pixel could determine the size of the

image, the resolution could determine the image

quality. Intensity was the number of colors contained

in an image. Intensity also has many terms such as:

256 colors, black and white (black & white), high

color, grayscale, and 16 million colors (true color).

Analysis of LSB Algorithm Modiﬁcation with Bit Inverse and Insertion based on Length of Message

523

The maximum number of colors in an image

depends on the file type (extension). File types with a

jpg extension could accommodate a maximum color

of 16 million colors, files with a .gif extension could

accommodate a maximum color of 265 colors

(Prabowo, Abdullah and Manik, 2018).

2.3 Image RGB

Color was identified in three-dimensional color

space, For types of hardweare-oriented color space

include RGB (Red Green Blue), CMY (Cyan

Magenta Yellow) and YIQ, while user-oriented color

space types include HLS (Hue Saturation

Luminance), HCV, HSV (Hue Saturation value),

HSB, MTM, CIE-LAB, and CIELUV .Most image

file formats (JPEG, BMP, GIF) use the RGB color

space. RGB color space was defined based on the

values of the axes R, G and B.

There were millions of colors in this nature. If

calculated based on variations in the value of R, G and

B that were owned by a color, then there would be

256 x 256 x 256 pieces of color. Each value of R, G

and B which was owned by a color varies between 0-

255. (Karma, 2020).

2.4 Image Quality Testing

Peak Signal to Noise Ratio (PSNR) and Mean

Squwere Error (MSE) was the most common

parameter used to measure the image quality testing

(stegoimage). PSNR told the similarity between the

original image and the image from the insertion and

was the opposite of MSE which was the damage value

of a stego-image. Mathematically PSNR in

formulated (1) and MSE in formulated (2)

𝑃𝑆𝑁𝑅 𝑑𝐵10𝑙𝑜𝑔





𝑃



𝑀𝑆𝐸



(1)

𝑀𝑆𝐸 

𝑅𝐶

𝑋

,

𝑋′

,















(2)

Which was:

P = maximum pixel value,

R = number of pixel in each row,

C = number of pixel in each column,

i ,j = row and column numbers,

i,j

= original image

X’

I,j

= stego-image.

Analysis obtained results showed that PSNR was

reduced when secret information size was increased

because of more pixel in cover image was changed

(more Noise) (Shehab and Abdulkadhim, 2018).

2.5 Linear Congruential Generator

(LCG) Algorithm

Linear Congruential Generator (LCG) algotrithm was

a simple pseudo-random number generator which

used to determine the next random number based-on

previously generated one. This approach had a

potential repetition on the generated numbers

whenever the numbers selected as parameter values

were not appropriately chosen (Sitompul et al, 2018),

Mathematically LCG will shown in formulated (3):

Xn = (a (Xn-1) + b) mod m

(3)

Where:

Xn-1 = previous random number

X = i-th random number

A = multiplier constant

B = constant increase (increase)

M = modulus constant

The period of LCG algorithm was not greater than

the modulus (m), the modulus was the maximum

threshold for randomizing numbers. So, the larger the

size of the container image, the more messages could

be inserted. (Hernandes, et al, 2019).

3 METHODOLOGY

The two LSB modifications have something in

common where the modifications are made based on

input. The first modification of the LSB is carried out

based on the container image, especially on bits 2 and

3 of the LSB. While the second modification is

carried out based on the length of the message to be

inserted into the image. The difference between the

two modifications is the position of pixel that are

inserted message, and the number of bits that are

inserted on insertion process.

3.1 Position of Pixel

In LSB modification with bit inverse, message

insertion is done on successive image pixels

(resulting in a decrease in image quality in a certain

area). LSB modification based on the message length,

the insertion is done at random pixels using Linear

Congruential Generator (LCG) Algorithm where the

constant modulus value is the number of pixels in an

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

524

image and the multiplier value with an increment of

random numbers between 1-100.

Sitompul et al (2018) have conducted a study on

random message insertion with a smaller MSE yield

and a larger PSNR. Figure 1 is an illustration of the

pixel position inserted with the message:

(a) (b)

Figure 1: Illustration of the pixel position inserted with the

message on LSB Modification with Bit Inverse (a) and LSB

modification with insertion based on the length of the

message (b).

From Figure 1, it is clear that the sequential

insertion of messages makes a decrease in image

quality in a certain area whereas random insertion

results in only a slight decrease in image quality

(inconspicuous and easily recognized by people

without access rights.

3.2 Insertion Process

3.2.1 LSB Modification with Bit Inverse

First, insert the message using the LSB algorithm then

the prepared message is in the inverse bit. If bits 2 and

3 LSB are the same then inverse is not required. The

insertion process is carried out in the following

stages: First, change the message and cover image to

binary, if the message bit is 010 and the image pixel

is:

01010111 01010011 01010111

then the message bit in the last binary of each pixel

layer is inserted or position 1 LSB:

01010110 01010011 01010110

After the insertion is complete, the 2

and 3

bits will

be visible. If the bits are the same, the message (bit 1

LSB) will not change, whereas if the bits are different,

the message (bit 1 LSB) will be inverse, like:

01010110

01010010 01010110

After checking each pixel where the message has

been inserted, the cover image binary will be

converted back into an image.

3.2.2 LSB Modification with Insertion Based

on Message Length

The first step in the modification is to calculate the

length of the message then convert the message into

binary form before the inverse binary message is

carried out. The length of the message is modulo with

2, if the result of modulo is 0 then 2 message bits will

be inserted at each pixel layer, whereas if the result of

modulo is 1 then 1 message bit will be inserted. After

the message has been processed, determine the pixels

to be inserted using the LCG algorithm, according to

the results of Zyaraa (2017) research use Linear

Congruential Generator (LCG) Algorithm will

increase the security of stegoimage.

Then convert the pixels into binary form. An

example of this modification if the result of modulo

is 0, the message bit is 101 and the binary of the

selected image pixel is:

01010011 01010011 01010011

The message will be inverted first to 010 then

insert 2 bits at once in each pixel layer at positions 1st

and 2nd LSB, then the insertion results are:

01010001 01010010 01010011

Meanwhile, if the result of the message modulo is

1 then insertion of 1 bit will be carried out in each

pixel layer at position 1

LSB so, the insertion result

will be:

01010010 01010011 01010010



After all insertion is complete, the binary of the

container image will be changed again and the image

will turn into a composite image and secret messages.

4 TESTING RESULT

In the experiment, the message used in the form of

text or images must be determined before inserting

the message. The requirement in testing is that the

message length cannot be greater than the cover

image (the resolution of the flax image used is 512 x

512, with lena (i) baboon (ii) and pepper (iii)). Image

2 is the image used in the study:

Analysis of LSB Algorithm Modiﬁcation with Bit Inverse and Insertion based on Length of Message

525

(

)

(ii)

(

iii

)

Figure 2: Cover Image.

4.1 Text Message Insertion

Following are the results of inserting a text message

with a character length of 2.110 and a number of bits

of 16.880. Modification of the LSB with insertion

based on the length of the message insertion is done

as much as 2 bits per byte, the second experiment was

carried out with a character length of 2.109 and a

number of bits of 16.872, where the LSB

modification with substitution based on the length of

the message insertion is 1 bit per byte. Then the

message is tested on the three cover images. In this

study, a system where after the message and cover

image are entered, it will automatically generate

MSE, PSNR, and insertion time has been developed

as in Figure 3:

Figure 3: Encoding process for text substituion in system.

MSE, PSNR and running time from figure 3 will

collect and analys, there would be a reduction in

image quality and This could be seen from the Mean

Squared Error (MSE), so the higher the MSE in an

image, the lower the quality of the resulting image

and the lower the Peak Signal to Noise Ratio (PSNR),

the image the resulting would resemble the original

image. The results of the MSE and PSNR results can

be seen in Table 1:

Table 1: Comparison of stegoimage Quality of Text

Message Insertion.

Based on Table 1, it can be seen that if each

experiment LSB modification with subtitution based

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

526

on mesaasge length is better than LSB modification

with bit inverse, it means that the use of Linear

Congruential Generator (LCG) Algorithm has a very

significant impact on the development carried out.

Furthermore, Table 2 also shows the time required to

insert the message.

Table 2: Comparison of Encoding Time on Text Message

Insertions.

From Table 2, it can be seen that the time required

for LSB modification with substitution based on the

length of the message looks longer because the

modifications made must do pixel randomization

before message insertion. Meanwhile, LSB

modification with bit inverse does not need to

randomize pixels and only inserts so that it takes less

time.

4.2 Image Insertion

Image insertion is not much different from text

insertion, so that in the experiment, an image length

of 100 and a number of bits 240000 where LSB

modification by insertion according to the length of

the message was carried out 2 bit each layer pixel.

The second experiment was carried out with an image

length of 105 and the number of bits 264600, where

LSB modification by insertion according to the length

of the message was carried out 1 bit each layer pixel.

Then the message is tested on the three cover

images. In this study, a system where after the

message and cover image are entered, it will

automatically generate MSE, PSNR, and insertion

time has been developed as in Figure 4:

Figure 4: Encoding process for image insertion in system.

The comparison of stegoimage quality as

evidenced by calculating mse and psnr will be shown

in Table 3.

Table 3: Comparison of stegoimage Quality of Image

Message Insertion.

From Table 3, it can be seen that the quality of the

message inserted 1 bit has a higher PSNR. This

happens because the resulting image reduction is

quite small, only 1 bit compared to 2-bit insertion

which gives more image degradation. Even so, the

LSB modification substitution based on the length of

the message with are still better than the LSB

modification with bit inverse. The time will be shown

in Table 4.

Analysis of LSB Algorithm Modiﬁcation with Bit Inverse and Insertion based on Length of Message

527

Table 4: Comparison of Encoding Time on Image Message

Insertion.

The time is not much different from the insertion

of text, only a little longer (because the modifications

made must randomize the pixels before inserting the

message). Meanwhile, LSB modification with bit

inverse does not need to randomize the embedded

pixels so that it takes less time.

4.3 Result Analysis

Based on the research that has been carried out, the

modification made by the author with the insertion

based on the length of the message provides better

stegoimage quality to LSB modification with the

inverse bit. This is evidenced by the average Peak

Signal to Noise Ratio (PSNR) as a variable which

states the similarity between the original image and

the higher stegoimage, both at the time of text

message insertion (66.29 dB> 61.8 dB) or image

insertion (54.20 dB> 50.01 dB). In the use of

steganography techniques, comparison of image

quality before and after message insertion is very

important. If a stegoimage file has a large enough file

size and the image quality is too bad, other people

who know about the use of steganography techniques

will be suspicious of the image.

The LSB modification by insertion based on

message length has the advantage of stegoimage

quality. This is because the used of Linear

Congruential Generator (LCG) Algorithm in

randomizing the pixels to be inserted into the

message. From those results, it can be concluded that

the developed modifications were very satisfactory as

seen from the higher quality of the stegoimage. While

deficiencies were found only during processing,

however, they are still within reasonable limits.

5 CONCLUSION

The use of the LCG algorithm in LSB modification

with message length insertion has a very significant

impact (as evidenced by the higher PSNR value than

LSB modification with the inverse bit insertion) both

at text message insertion (66.29 dB> 61.8 dB) and

image insertion (54.20 dB) > 50.01 dB). For further

research, the authors plan to compress the stegoimage

to facilitate the sending of images containing secret

messages. Experiments carried out on the condition

that the decoded stegoimage must not delete

messages with exactly the same order as the original

message. Furthermore, a given stegoimage tolerance

to noise will be tested.

REFERENCES

Al-Aidroos, N. M., Bahamish, H. A., 2019, Image

Steganography Based on LSB Matching and Image

Enlargement, First International Conference of

Intelligent Computing and Engineering (ICOICE).

IEEE

Bhardwaj, R., Sharma, V., 2016., Image Steganography

Based on Complemented Message and Inverted bit LSB

Insertion, Procedia Computer Science, vol. 93, pp.832-

838.

Gunawan, I., Sumarno., 2018, Penggunaan Algoritma

Kriptografi Steganografi Least Significant Bit Untuk

Pengamanan Pesan Teks dan Data Video, J-SAKTI,

vol.2, pp.57-65.

Hernandes, A., Hartini, Sartika, D., 2019. Steganografi

Citra Menggunakan Metode Least Significant Bit

(LSB) Dan Linear Congruential Generator (LCG),

JATWASI, vol. 5, pp. 137-150.

Karma, I., G., M, 2020. Determination and Measurement of

Color Dwassimilarity. International Journal of

Engineering and Emerging Technology, vol. 5(1)

pp.67-71.

Prabowo, D.A., Abdullah, D., Manik, A., 2018, Deteksi

Dan Perhitungan Objek Berdasarkan Warna

Menggunakan Color Object Tracking, Journal

Pseudocode, vol.5, pp.85-91.

Sari, R.D., Siahaan, A. P. U., 2018, Least Significant Bit

Comparison between 1-bit and 2-bit Insertion,

International Journal for Innovative Research in Multi

Disciplinary Field, vol. 4(10), pp. 110-113.

Shehab, J. N., Abdulkhadim, H. A., 2018. Image

Steganography Based on Least Significant Bit (LSB)

and 4-Dimensional Lu and Liu Chaotic System,

International Conference on Advanced Science and

Engineering (ICOASE), pp. 274 - 279. IEEE

Shtayt, B.A., Zakaria, N. H. B., Harun, N. H., 2020, A

Comprehensive Review of Spatial Domain-Based

Image Steganography Techniques.International

Journal of Advanced Science and Technology, vol. 29

(7), pp. 13311 -13329.

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

528

Singh, A., Chauhan, M., Shukla, S., 2018, Comparison of

LSB and Proposed Modified DWT Algorithm for

Image Steganography. International Conference on

Advances in Computing, Communication Control and

Networking (ICACCCN2018) pp. 889-893. IEEE

Sitompul, O.S., Naibaho, F.R., Situmorang, Z., Nababaan,

E.B., 2018, Steganography with Highly Random Linear

Congruential Generator for Security Enhancement.

International Conference on Informatics and

Computing (ICIC). IEEE.

Zyaraa. A. H., 2017, Suggested m.thod for encryption and

hiding image using LCG and LSB, Journal of

University of Babylon for Pure and Applied Sciences

pp.1-16.

Analysis of LSB Algorithm Modiﬁcation with Bit Inverse and Insertion based on Length of Message

529