Image Steganography Method for Securing Multiple Images using LSB – GA

Image steganography is renowned for its �exibility and frequency in the current scenario. This article presents a combination of LSB (Least Signi�cant Bit technique) and GA (Genetic Algorithm) to hide various forms of data – text and images. A novel combination of steganography techniques is used to transmit complete secret data. Also, multiple color images are embedded in a single color image, showing increased secret data holding capacity. Furthermore, the application of GA assures increased security. The excellence of the stego image and retrieved images are evaluated in terms of Peak Signal to Noise Ratio (PSNR), Structural Similarity Index Metric (SSIM), Correlation coe�cient, and Bit Error Rate (BER). Hence, this article is targeted to transmit multiple color images with no loss and improved security using GA and we achieved SSIM between 0.996 to 1.


Introduction
Steganography is de ned as the skill of concealing information by embedding messages within other messages.It is more likely associated with data hidden within another data in an electronic le.Steganography is originated from the Greek words "Steganos," meaning covered or secret, and "Graphia" meaning writing or drawing [1].Thus, steganography is hidden writing, consisting of invisible ink on paper or copyright information hidden in a le.Steganography can be classi ed as Audio steganography, Image steganography, Video steganography [2].Among these, Image steganography is often preferred in this digital world.Image steganography techniques can be apportioned into two major classes: Spatial domain embedding and Transform or Frequency domain embedding.Spatial domain techniques entrench information in the intensity of the original image pixels directly.The commonly least signi cant bit (LSB) method is used where it replaces the least signi cant bit of the original pixel with the message bit.There are many versions of spatial steganography, and all immediately modify some bits in the image pixel values in concealing data.The least signi cant bit (LSB) based steganography is one of the most straightforward techniques to conceal secret information in the LSBs of pixel values without creating any noticeable distortions.Human eyes cannot predict the changes in the value of the LSB [17].
Generally, we mention the transform domain as the frequency domain, where images are rst transformed, and then messages are embedded in the image.The discrete cosine transformation (DCT) technique is used in JPEG images to achieve compression.Discrete Wavelet Transform (DWT) is another transform that gives four sub-bands of the image.This is a more complicated method of concealing data in an image.Transform domain embedding is a category of embedding techniques for which several algorithms have been proposed.The method of embedding data in a signal's frequency domain is far more powerful than embedding principles in the time domain.
The transform domain is where most solid steganographic systems function today.This technique has the advantage of hiding information in portions of the image that are less vulnerable to compression, cropping, and image processing than spatial domain techniques.All transform domain approaches are independent of image format, and they can be used to convert between lossless and lossy formats.
Usually, in the encoding or embedding process, the pixels or binary values of text or pixels of images are hidden using bitwise operations.These operations later few bits in an extensive cover and thus enable the secret data to be present in the cover image when transmitted.The alteration must not produce visible changes in the cover image.
The idea of hiding multiple color images in a single color image with increased security is achieved by applying the Genetic Algorithm (GA).However, GA changes the exact pixel orbit locations, complicating the extraction of the embedded secret message.Hence, this article intends to provide an improved level of security to stego images carrying a high amount of secret data [8].
The article is organized as follows.In Section 2, related work of the article is summarized.Section 3 explains the proposed algorithm, and the corresponding results and discussion are included in Section 4.
Finally, section 5 concludes this article.

Related Work
Singh et.al delivers an overview of image steganography, its applications, advantages and disadvantages, and various techniques used.Since the rise of the internet, where anyone can access personal information, security has become an essential factor in information technology and communication.Therefore, it becomes necessary to provide security to data so that no unauthorized person can access it [19].
Isha Kajal et al gives an overview of all the different techniques used in image steganography.Some of the existing essential image steganography techniques like Least Signi cant Bit (LSB), Pixel Value Differencing (PVD), and Modi ed Kekre Algorithm (MKA) etc.Here the various methods are compared as capacity versus quality.When the number of bits of the secret data is low, the stego image quality is high and vice versa [14].
Laskar et.al proposed a combination of steganography and cryptography.The data is rst encrypted using the transposition cipher method, and then the encrypted information is then embedded inside an image using the LSB insertion method.This results in the enhanced security of the data embedded and also provides a higher capacity.It satis es the capacity, security, and robustness requirements for secure data transmission over an open channel [26].
Using the Discrete Wavelet Transform approach, Baby et al suggested a data securing technique that hide numerous color images into a single color image.First, the cover image is split up into R, G and B planes.Then, secret images are embedded into these R, G, and B planes.Finally, the cover image and the secret images are N-level decomposed, and some frequency components are merged.The stego picture is then used to extract secret images.Therefore, the stego image obtained has less perceptible changes than the original image with high overall security [7].
Mandal et al. proposed a Genetic Algorithm based color image authentication/data hiding technique through a steganographic approach.It is termed Genetic-Algorithm-based Color Images with Steganography (GASCI).To construct the stego image, the message/image is embedded on a spatial domain onto the rightmost three bits of each byte using 3 x 3 masks derived from the source picture in row-major order.To improve security, New Generation, Crossover, and Mutation are used.During decoding, the procedure is reversed.Various statistical metrics computed are compared to existing Yu et Rohit et al. introduced various noise models and removed them with different types of lters.The noise attacks the steganographic image during image acquisition or transmission.Furthermore, this paper presents the various ltering techniques that can be applied to de-noise the images.Therefore, getting an e cient method of removing noise from the images is a great challenge [25].Kamboj et al. presents a complete and quantitative analysis of noise models available in digital images.Since noise is challenging to remove from digital images, it is necessary to understand the various noise models essential in studying image de-noising techniques.Noise models are also designed by probability density function using mean, variance, and gray levels in digital images [17].
Rehana et al. provide the best approach for secure data hiding and transmission over Networks using LSB-based steganography with Genetic Algorithm (GA) and Visual Cryptography (VC).The system here encodes the secret message in the least signi cant bits of the cover image so termed as a stego image by using a secret key.Genetic Algorithm and Visual Cryptography have been used for enhancing security.First, the pixel placement of the stego picture is modi ed using a Genetic Algorithm, which is another protection lock for the secret message and image.The detection of this is complex.Visual Cryptography is further used to encrypt the modi ed pixel image by breaking it into two shares based on a speci c threshold.Later those encrypted shares and the secret key is separately sent to others using Network Socket Programming.Finally, the user who received the secret shares must do the reverse process to retrieve the secret key's image and message.The implementation is done in the java platform, which shows that the proposed system is highly secure and reliable [24].Khamrui et al. proposed a Genetic Algorithm based steganographic technique in the frequency domain using discrete cosine transform.To produce four frequency components, a 2 x 2 sub mask of the original image is obtained in row signi cant order and Discrete Cosine Transformation is done.Except for the rst, each modi ed coe cient contains two bits of the authenticating image.The second and third places from LSB are chosen for embedding in the transform domain in each coe cient.The reverse transform is used to create the Stego sub intermediate image.As the starting population, a sub mask from this intermediate image is used.New Generation followed by crossover is applied to the initial population to enhance a layer of security.The New Generation has grown up with the original population.The rightmost three bits of each byte are taken, and a three-step bitwise XOR is used to construct a triangular form.Each intermediate step's rst bit is used as the output, and crossover is done on two consecutive pixels by swapping the two LSB bits of two consecutive bytes.The concealed image dimension is embedded rst, followed by the content.During decoding, the procedure is reversed.In the stego image, the suggested technique achieves great image quality, PSNR, and embedding capacity [4].
Ahmed et al. brie y outlined how images can also be embedded as text steganography.Adding one image in each bit plane up to seven images has been dealt with as an actual example.It is noticeable that embedding more than one image in an image loses its resolution.However, we can go for this proposed technique where image quality is not needed, and the number of images is essential [10].

Mehra et al. deal with a secure transmission in the wireless medium through the technique of Least
Signi cant Bit (LSB) based steganography using Genetic Algorithm (GA) along with Visual Cryptography (VC).It initiates with the original message, which is converted into ciphertext by using the secret key and then hidden into the LSB of the original image.For enhancing the security during transmission Genetic Algorithm and Visual Cryptography has been used [21].The Genetic Algorithm essential function is to modify the pixel location of the stego image and detect this message is complicated.Encryption of the visual information is conducted by Visual Cryptography.It breaks the image into two shares based on a threshold.The proposed system is experimented with by performing steganalysis and analyzing the parameters like Mean Squared Error (MSE) and Peak Signal to Noise Ratio (PSNR) [13].
Bhattacharya et al. proposed that there is a large variety of Steganography techniques for hiding secret information in images.The least signi cant bit (LSB) insertion method is the most general approach for embedding messages in a spatial domain image is the least signi cant bit (LSB) insertion method.They focussed on combining various photographs into a single cover image.Three images can be placed between the least signi cant bit (LSB) and a moderately signi cant bit (MSB) in a single cover image with three separate color channels (RGB) (MdSB).Three separate passwords are used for three different color channels, which are later translated into keys.Three separate passwords yield three different keys with 64 components each.The key is made up of the position values of a colored channel of the cover image, in which information from a grey image is incorporated using the variable bit replacement (VBR) approach.Three separate channels are combined to generate a stego image, and grey images are extracted in the same way from the stego image.The change from password to key, as well as the reliance on key, increases security.Another signi cant feature is that the same stego image might have multiple meanings for different receivers [30].This section introduces the two primary techniques used in this paper.

LSB Technique
This is the most basic and essential image Steganographic Technique embedding technique.Data can be buried in the least signi cant portions of the cover image using this technique, and the hidden image in the cover le remains undetectable to the naked eye.This method can be used to hide images that are 24-bit, 8-bit, or grayscale.Here, each pixel's two least signi cant bits are replaced with the secret message bit until the message end.For example, when using a 24-bit image, one can store six bits in each pixel by changing at least two bits of each red, green, and blue color component [18].
For example, a twenty-four bit can be as follows: Although the number is embedded into the rst 4 bytes of the cover, only the two underlined bits in each byte need to be changed according to the embedded message.To hide a secret message utilizing the maximum cover size, only half of the image bits will need to be updated on average.There are 256 intensities for an 8-bit representation of each primary color, so changing the LSB of pixel results in minor changes in the intensity of the colors [23].

Genetic Algorithm (GA)
The genetic algorithm-based steganography method incorporates any image embedding technique to hide the data in an image with improved security.The genetic algorithm is a problem-solving method based on natural selection, the driving force behind biological evolution.The genetic algorithm repeatedly modi es a population of individual solutions.The genetic algorithm randomly selects parents from the current population at each phase and uses them to produce the following generation of children.Thus, over successive generations, the population evolves towards an optimal solution.The genetic algorithm can be used to handle a variety of optimization issues that aren't well suited to traditional optimization algorithms, such as problems with a discontinuous, non-differentiable, stochastic, or highly nonlinear objective function.In image steganography, GA is concerned with improving security rather than optimization.
The genetic algorithm can be applied to solve issues involving mixed-integer programming, in which speci c components must be integer-valued.To build the next generation from the current population, the genetic algorithm employs three sorts of rules at each phase: Selection rules determine which individuals, referred to as parents, will contribute to the next generation's population.
Crossover rules combine the children of two parents to create the following generation.
Mutation rules apply random modi cations to individual parents to generate children.
This causes considerable changes in the data and hence protects the information hidden using steganography.In spatial domain techniques of steganography, if a genetic algorithm is used to modify the stego object values without causing visual distortions slightly, it gives the stego object a better security level.So, in this work, we combine the LSB or spatial domain technique of steganography and a genetic algorithm to secure secret data transmission [28].

Proposed Work
Boyat et al. concludes that spatial techniques (LSB) provide more hiding space than the transform techniques.Also, in transform domain techniques, only the approximation coe cients of the secret data are hidden and retrieved.That may offer robustness to attacks but cannot make the complete secret to be transmitted and retrieved.Hence, we implement a method in the spatial domain to hide and extract secret text.The image of unusual sizes and multiple images are hidden in a single RGB color image without losing any part of the secret data.Those applications that require and aim to transmit complete secret data can opt for our image steganography method [3].
The data (text, image(s)) required to be hidden in the image is chosen and converted into an 8-bit representation.Text is represented using ASCII code.Images are described in the grayscale format, and the RGB color image is hidden by slicing it into its three constitute planes.To increase the security level in the spatial domain, we propose the application of GA.This genetic modi cation of the bits in the stego image is done only in the LSBs.Hence visually, the stego image is least affected.Moreover, the results are entirely uncorrelated with the original embedded data if the extraction is done without reversing the GA.Hence this proves to be better security for spatial domain image steganography.
GA is applied in three steps: initial population selection, next-generation formation, the crossover of bit locations, and nally, mutation.As all these are customized according to the programmer, it is challenging to track embedded secret data.Therefore, the data is embedded in the image using -LSB embedding techniques separately.This yields a stego-image that contains the secret data invisibly.The received stego image has rst undergone a reversal of GA.Then, proper decoding of the obtained stego image is done to extract the embedded secret information.To evaluate the performance of each algorithm, a few parameters are considered for the stego-image and original cover image, secret data, and recovered data [29].BER and Correlation coe cient is determined between the recovered image and the original secret image in the above parameters.The other parameters, PSNR, SSIM, MSE, Correlation coe cient, are measured between the original cover image and the obtained stego image.Our results prove that there is a high capacity of embedding and improved security due to the application of GA.Furthermore, SSIM shows that the stego image is entirely similar to the original cover image.
When one color image is to be hidden in the RGB color cover image, R, G, B planes of the secret image must be sliced initially.Each of these planes must be hidden in the pixels of the cover image without overlapping.This requires the cover to be at least three times bigger in size than the secret image.When We propose this work in a linear sequential manner so that it remains invisible.The visual appearance of the stego image does not give way to predict the existence of secret data.Even the part of the stego image which contains the secret message is known only to the programmer.Hence, even in the LSB technique that allows holding more data but is less secure, it is possible to transmit reliably when GA is applied.Initial articles on LSB image steganography or image watermarking show ways to hide black and white or binary text or images only.Also, we used our techniques to JPEG images that are already compressed, and hence the problem of compression does not erase the embedded secret information [6].
One of the advantages of image steganography is that the size of the cover image remains the same after embedding.So, this does not cause suspicion when transmitted, as it looks and weighs the same as that of the cover image.However, after extracting the secret data, the cover image is not restored from the stego image.So, this image steganography can be improvised by making the cover image also restorable to its original state, which is called reversible steganography.
First, we hide the text in a color image, retrieve it and calculate the parameters.Finally, we apply the Genetic Algorithm (GA) to improve the security of the stego image that is transmitted.By using GA, there is a change in the ordering of hidden bits in every pixel of the stego image.This may seem to decrease the PSNR of the stego image.Still, when an intruder extracts the LSBs by guesswork on embedding algorithm, it produces unintelligible secret data, which is useless.Hence, GA protects the spatial domain steganography technique.Our GA can be applied to any stego image, and it is completely reversible at the receiver end.Therefore, there is no damage caused to the secret data contained in the stego image [12].Where MAX is the maximum value of pixels (255 for grayscale images), MSE is the mean square error between the original and stego images.Greater PSNR values indicate the best quality [5].It is expressed in decibels (dB).

Mean Squared Error (Mse)
MSE describes the mean square error between the original and stego images [9].For example, the below equation expresses MSE.
where is original pixel and is stego pixel.

Structural Similarity Index Metric (Ssim)
SSIM is an objective image excellence metric and is superior to traditional measures such as MSE and PSNR.PSNR estimates the perceived errors, whereas SSIM considers image degradation as perceived change in structural information.Structural information is the concept that the pixels have strong interdependencies, especially when they are spatially close.These dependencies carry valuable information about the structure of the objects in the visual scene [11].The equation gives the SSIM, Where C1 = (kL) and C2 = (kL) are two constants used to avoid null denominators.
L is the dynamic range of the pixel values (typically, this is two bits per pixel -1).= 0.01 and = 0.03 by default.The dynamic range of SSIM is between -1 and 1.A maximum value of 1 will be obtained for identical images.
It can be written as the product of three terms: M1, M2, and M3 given as, where , M1 indicates luminance distortion, M2 indicates contrast distortion, and M3 shows structural distortion.

Bit Error Rate (Ber)
The Bit Error Rate (BER) is the number of bit errors per unit time.The BER is the number of bit errors divided by the total number of transferred bits during a studied time interval [20].Thus, BER is a unitless performance measure.

Correlation Coe cient
The Correlation coe cient formula determines the relationship between data.
where =mean(A), and =mean(B), A, and B are two-dimensional matrices of the same size (m x n).
STEP-2: Read the message to be hidden -RGB image.
STEP-3: Slice the R, G, B planes of the message.
STEP-4: Each plane is hidden in the cover image one by one.
STEP-5: Repeat STEP-2 to STEP-4 for multiple images up to the size of the cover image.
STEP-6: Display and save the stego image obtained.

Genetic Algorithm
Now we apply a Genetic Algorithm to increase the level of security of the stego image.This GA applies to any embedding technique mentioned above.Here we apply it to a stego image containing multiple color secret images using steganography as in section 4.5.The genetic algorithm employs three main functions to modify each pixel in the stego image.When the embedded bits in each pixel are jumbled up, it becomes di cult for the third-party intruder to extract the secret data.Each function of GA is customizable according to the idea of its programmer.Here presents the way of application of GA in our work:

Selection of initial population
STEP-1: Slice R, G, B planes of the stego image.
STEP-2: Choose B plane STEP-3: For a few rows in each column, extract the LSBs which were embedded.
STEP-4: Continue STEP-3 and form an initial population matrix.
Replacing in the plane STEP-1: Read elements from the next generation matrix one by one.
STEP-2: From each element, replace the bits taken from the respective B plane elements.

Crossover
Take each element from the R and G planes.These are the parents.
STEP-3: Exchange embedded bit positions between parents and children.
STEP-4: Replace the crossed-over children in their respective planes.

Mutation
STEP-1: Make a random change in the pixel value throughout the entire stego image.

Extraction algorithm
STEP-1: Read the stego image.
STEP-2: Reverse the GA in the reverse order -Mutation cancellation, Cross over recovery, Generation reversal.
STEP-2: Extract the separate R, G, and B planes of the message from the correct pixels from where it is hidden in the stego image.
STEP-3: Combine the R, G, B planes obtained separately in the previous step to get the single RGB image.
STEP-4: Repeat STEP-2 and STEP-3 up to the size of the cover image to obtain all the color images.
STEP-5: Save the extracted data as an image.
STEP-6: Verify the correctness of the extracted data using BER and SSIM.
The keys used for next-generation formation are randomly generated.So, GA can be reversed if the same keys are known.These keys are also embedded in the bottom part of the nal stego image and transmitted.After reception of the stego image, the receiver must rst go to silent the mutation.Then, they must reverse the crossover and nally reverse the new generation.Then the normal extraction process should be done to obtain the secret data [27].

Results
Spatial domain technique -LSB gives more space to hide.So, multiple images, even multiple color images, can be embedded and retrieved successfully with no error.
Baby et al. [1] experiments color in color image embedding in the transform domain by applying DWT to cover and secret images.However, only the most signi cant coe cients of the secret images are hidden and retrieved.Hence, originality or accuracy is reduced in the extracted images.Our method ensures complete embedding and extracting of the secret images with 100% correlation and SSIM values.The only constraint is that the cover image should be large enough to hold all the secret color images chosen.
The following table shows the variation in the cover after embedding, i.e., the comparison between the unaltered cover image and the stego image.It is measured in terms of PSNR, MSE, SSIM, and Correlation coe cient.In the transform domain, the capacity of hiding is less than the spatial domain.Hence, only approximate coe cients are used to hide the images, and retrieving of the images is done using approximate coe cients only.Therefore the secret message is only 78.4% (average) as the original secret images.In our method, we obtained 100% similarity to the original secret image after extraction.In the DWT technique, PSNR values are high, which indicates that the transform domain hiding does not cause any changes to the pixel values of the cover image.On the other hand, in LSB hiding, a total of 8 bits are replaced for each color pixel.So, the PSNR values are less than the transform domain approach.However, steganography aims to hide the data invisibly.Therefore, the SSIM value is a better parameter to show the visual accuracy of the stego image and the extracted image.Hence the proposed technique of LSB proves to satisfy the needs of steganography in a better way.
GA is applied to increase the security of the LSB technique, as explained in the previous chapter.We include the results that show how the stego image affects the application of GA, the results when the extraction is done without reversing GA.When GA is reversed and extraction is done, we get the complete secret data without loss.Though there is a decrease in the PSNR value after applying the Genetic Algorithm, it is acceptable as the visual details are not affected.Moreover, GA ensures better protection to the stego image.Therefore, this genetically modi ed stego image is more secured.When received, GA must be reversed, and the extraction process is done.If extraction is done without reversing GA, the output appears as shown in the gure below.The recovered images are not proper, and hence GA ensures security.
The following gures show the GA reversed stego image and the subsequent extraction process.The parameters for the extracted images are tabulated below it.Thus, our proposed system combines the LSB spatial domain steganography technique features and a superior level of security using GA.Furthermore, the extraction process shows that complete secret data is extracted, which is evident through the SSIM values, which remain one.Stego image after application of GA

Conclusion
Extraction without reversing GA Extraction after reversing GA al. (2010) and Wang et al. (2001) data, demonstrating that the proposed GASCI produced improved PSNR results [16].
number 111 which binary representation is 01101111, is embedded into the least signi cant bits of this part of the image, and the result is as follows: The parameters are: Peak Signal to Noise Ratio (PSNR) Mean Squared Error (MSE) Structural Similarity Index Metric (SSIM) Correlation coe cient Bit Error Rate (BER)

STEP- 7 :
Calculate the parameters to assess the stego image quality.

Table - 1
: Parameters measured for LSB embedding of multiple color images in a single RGB cover image acceptable since, visually, there is no visible distortion in the stego image.MSE shows the squared values of the error between pixels of the cover image and the stego image.The correlation between cover and stego images is remarkably close to one indicating a higher correlation.SSIM values suggest that the stego images are on average 99.67% structurally like the cover images.Table-2: Parameters measured for LSB extraction of multiple color images from the stego imageFrom the previous tables, it can be understood that the proposed method to hide and extract multiple images inside a single color cover image retrieves the secret images as such without any loss.The table-5.10 shows the SSIM values obtained in the DWT technique.

Table - 4
: Comparison of PSNR values measured in DWT technique and LSB technique

Table - 5
: Parameters for stego image with GA

Table - 6
: Parameters for recovered secret images after reversing GA