Image steganography is a method of giving hidden data to a host image to transfer it in a secure manner. There are many techniques available for this. Choosing the proper algorithm is a necessary step. This paper presents a survey of popular image steganography techniques and also discusses their advantages, limitations and effectiveness in comparision. The basics of image steganography, factors for evaluating an image steganography are covered in the paper. Image steganography techniques can be classified according to various criteria. The classifications are also covered in this paper. Five major image steganography techniques have been covered in this paper. An indepth analysis of each technique, their merits and demerits and comparision of the techniques are done in this paper. Various steganalysis techniques has also been covered.
Image Steganography is in which data is hidden within an image file. The image selected for it is called as the cover-image and the image obtained after the steganography process is called the stego-image. Image steganography is of various types.
In image steganography an image is usually the carrier that holds the secret information. In the diagram below ,cover image shows the image which is used for hiding the secret data as a payload. The embedding technique is the algorithm which is useful for hiding the secret message inside the cover image with the help of stego key. The stego-key needs to be shared with both the ends. The ‘stego-image’ is the final output image that keeps the secret information hidden. There is also an extraction technique which is the used to recover the secret message from stego-image with the help of stego-key. The counter of steganography is steganalysis or an attack on steganography. “Steganalysis is an art and science of detecting the existence or recovering the secret message from stego images.”
Fig 1: Basic functioning of Image Steganography
While going through different papers and resources on image steganography it was important to know and understand on what basis they were formed and differenciated.
A image steganographic method is evaluated based on the following mainly
- Embedding capacity- What is the maximum embedding payload that can be achieved
- Visual image quality- To what extent are the stego-image and the cover image identical
- Security- How can a stegoimage protect itself from various steganalysis detection attacks
An ideal steganographic method should fulfill the above mentioned points. But in reality, high payload steganographic methods introduce the distortion artifacts in stegimages and that make it vulnerable to steganalysis. Also the steganographic techniques which has a good visual image quality suffer from issue of low payload. Therefore achieving high payload, good visual quality, and undetectability simultaneously is a challenge.
The image steganographic methods proposed can be classified based on various criteria.Understanding the classification will also help us in understanding the methods. Classifying them based on model technique they can again be divided into different types based on their implementation.In Fig 2a, they are divided based on embedding domain, ie. spatial domain and transform domain .The adaptive embedding method is also under the above division because it can be used in both spatial and transform domains. Mainly this classification is based on their general objectives or goals. 2b shows the second type of classification based on distinct steganographic techniques which are specifically designed to target coded formats of images, i.e. raw (BMP), compressed (JPEG2000) and encrypted image data (AES-advanced encryption standard). 2c is based on classification based upon steganographic techniques which are designed with respect to the format/type of secret data.This division of steganographic techniques are based on various secret data types.
Fig 2. Classification of Image steganography
3.DIFFERENT TYPES OF IMAGE STEGANOGRAPHY
3.1 Least Significant Bit Insertion
It is spatial domain image steganography. Spatial domain embedding methods are generally more popular than transform domain because of its easiness in embedding and also extraction process. But they tend to be less robust.
Least Significant Bit (LSB) insertion is the most widely known algorithm for image steganography . It involves modification of the LSB layer of image. In this technique,the message is stored in the LSB of the pixels which can be considered as random noise.Therefore, altering them does not have an obvious effect to the image.
This method changes the last few bits in a byte to encode a message which is especially useful in case of image, where the red, green, and blue values of each pixel is represented by eight bits ranging from 0 to 255 in decimal or 00000000 to 11111111 in binary.
For example changing the last two bits in a complete red pixel from 11111111 to 11111101 only changes the red value from 255 to 253, which in case of human eye create a nearly imperceptible change in color but it still allow us to encode data inside of the picture.
This technique works very good for media files where the slight change of byte values creates only very small imperceptible changes. But it is not the case for ASCII text where a single bit out of place will entirely change the character. The data hidden using LSB steganography is easy to detect if someone is searching for it.
Fig 3: An example of LSB Steganography
These are the review of some methods according LSB techniques. Chan and Chang  have made use of the simple LSB technique with optimal pixel adjustment to improve the quality of stegoimage. Zhang and Tang  in their paper suggested an improved LSB technique by selecting a set of cover image pixels at a time dynamically with the pseudo random number. They embedded n bits of secret data in every pixel of image using concept of modular division operations and addition. The n value is based on length of embedded bit stream of secret data. Both security and capacity are targeted in this technique.
Although LSB based methods are considered a simple way of information hiding and even very flexible to integrate with other methods, the main disadvantage is that the embedding capacity has a direct relation to visual quality of stego-image.
LSB Matching (LSBM) is a much improved method over LSB replacement. LSBM is very similar to the LSB algorithm discussed above. In case the bit of the confidential message is not equivalent to the LSB that come from the cover pixel, 1 is either randomly summed up or subtracted from the value of the cover pixel. As compared to LSB Replacement method it is hard to detect LSB matching. That is for LSBM, if the secret bit is not equal to the LSB of the given pixel, then ±1 is added randomly to the pixel while keeping the altered pixel value in the range of [0,255]. 
In LSB, only the LSB bit plane is altered as a result a structural asymmetry is easily visible to steganolysis algorithms like Chi-Squared attack. In LSBM the probability of increasing or decreasing for every modified pixel is the same. Therefore the asymmetry artifacts introduced by LSB replacement is avoided. So common approaches used to detect LSB replacement are totally ineffective at detecting the LSBM.
The main advantage of using LSB embedding is its simplicity and the difference is not visible to our naked eyes. But this technique also has a lot of disadvantages mainly like LSB encoding is extremely sensitive to any type of filtering or manipulation of the stegoimage. An attack made on the stegoimage is very likely to destroy the message. An attacker can easily remove the message by removing the entire LSB plane with very little change in the perceptual quality of the modified stego-image.
The main advantages of the LSB method are its higher hiding capacity means larger amount of information can be embedded in an image. And also the degradation of the original image is not easy.
Main drawbacks of LSB method are that the embedded data can be destroyed by simple attacks, therefore it offers only moderate security. Low robustness is a drawback. It is also vulnerable to attacks like formatting and image changes (like change from .GIF to .JPEG).
There are many uses of LSB mainly storing passwords and other confidential information,covert communication of sensitive data, being widely used to transfer illegal content. There are also speculated use in terrorist activity
3.2 Redundant Pattern Encoding
Redundant pattern encoding is kind of similar to spread spectrum technique. In this technique the message is scattered all over the image,on the basis on algorithm. Therefore the image based on this technique can not be cropped or rotated. Multiple smaller images with redundancy increase the chance of recovering, even in cases where the stegaimage is manipulated.
In case of redundant pattern encoding, a small message may be painted again and again many times over a image so that when the the stegoimage is cropped there is a high chance that the watermark can still be read.
Fig 4. Image ‘painted’ with watermark
Patchwork is a hybrid domain steganography that makes use of redundant pattern encoding. Patchwork is a statistical technique that uses redundant pattern encoding to embed a message within an image. The algorithm adds up redundancy on the hidden information, after which it is scattered through the image.
A pseudorandom generator is used in selecting two areas of the image like patch A and patch B. All the pixels in patch A are lightened where as the pixels in patch B are darkened. In simple words the intensities of the pixels in the one patch are increased by a constant value, where as the pixels of the other patch are decreased with the same constant value. The contrast changes in this patch subset encodes one bit. These changes are usually small and imperceptible, while the average luminosity reamins the same.
This approach can be used independent of the host image and proves to be robust since the hidden message may continue with transform between lossless lossy and compression. Patchwork is an application of redundant pattern encoding.
The disadvantage of the patchwork approach is that only one bit is embedded. By first dividing the image into sub-images and applying the embedding to each of them ,more bits can be embedded. The advantage of using this method is that the secret message is distributed over the entire image, so in case one patch is destroyed, the others can still survive. This depend on the message size because the message can be repeated through the image only if it is small enough. If the message is too big the embedding can be done only once. The patchwork approach is used independent of the host image and proves to be quite robust as the hidden message can survive conversion between lossy and lossless compression.
3.3 Encrypt and Scatter
This method hides the message as white noise. It is used mostly in image steganography. An example for it is White Noise Storm which uses spread spectrum and frequency hopping. Previous window size and data channel are used for generating a random number. Message is scattered on all the eight channels with in this random number.Each of these channels interlaces, rotates and swaps with every other channel. Single channel represent one bit and therefore there are many unaffected bits in each channel. 
In this method it is very complex to draw out the real message from stegoimage. This method is more secure when compared to LSB as it needs both algorithm and key to decode the bit message from stegoimage. Therefore users prefer this method for its security as it requires both algorithm and key despite the steganimage. This techniques just like LSB has de merits like image degradation in terms of image processing and compression.
Some people prefer this technique due to the considerable amount of extra effort that a person without the stego-key and algorithm would have to go through to extract the message. Even though White Noise Storm provides gives extra security against extraction of message it is just as susceptible as LSB to image degradation due to image processing.
3.4 Algorithms and transformation
LSB modification technique is not useful if any type of compression has been done on the resultant stego image like for example JPEG, GIF. JPEG images use the discrete cosine transform for compression. Since the cosine values cannot be calculated exactly, DCT is a lossy compression transform.  And also repeated calculations with limited precision numbers bring rounding errors into the final answer. Variances between original data value and restored data value is dependent on the method used to calculate the DCT.
Fig 5. Transform
A more complicated way of hiding secret inside a image is with the use and modifications of discrete cosine transformations. Discrete cosine transformations is used by the JPEG compression algorithm for transforming successive 8 x 8 pixel blocks of the image, into 64 DCT coefficients each. Each DCT coefficient F(u, v) of an 8 x 8 block of image pixels f(x, y) is usually shown by:
After calculating the coefficients, the following quantizing operation is performed:
Where Q(u, v) is a 64-element quantization table.
A simple pseudo-code algorithm to hide a message inside a JPEG image could look like this :
Input: message, cover image
Output: steganographic image containing message while data left to embed do
get next DCT coefficient from cover image if DCT 6= 0 and DCT 6= 1 then
get next LSB from message
replace DCT LSB with message bit
insert DCT into steganographic image
Initially there was a notion that steganography won’t be possible with JPEG images, since they make use of lossy compression which results in parts of the image data being altered. One of the main features of steganography is that the information is hidden in the redundant bits of an object. Since redundant bits are left out when using JPEG it was thought that the hidden message would get destroyed. Even if the message could be kept intact somehow it will be difficult to embed the message without any changes being noticed. This is mostly due to the harsh compression being applied. 
The properties of the compression algorithm have been exploited for developing a steganographic algorithm for JPEGs. One of the properties of JPEG that is exploited to make the changes to the image to make it invisible to the human eye. During the DCT transformation phase of compression algorithm, in the coefficient data rounding errors occur that are not easily noticeable.Though this property is what classifies the algorithm as being lossy, the very same property can also be used to hide messages.
It is neither feasible nor possible to embed information in an image that uses lossy compression, since the compression would destroy all information in the process. Thus it is important to recognize that the JPEG compression algorithm is actually divided into lossy and lossless stages. The DCT and the quantization phase form part of the lossy stage, while the Huffman encoding used to further compress the data is lossless. Steganography can take place between these two stages. Using the same principles of LSB insertion the message can be embedded into the least significant bits of the coefficients before applying the Huffman encoding. By embedding the information at this stage, in the transform domain, it is extremely difficult to detect, since it is not in the visual domain.
3.5 Masking and Filtering
Masking and Filtering is a steganography technique that is used on gray- scale images. Masking and filtering method works better with 24 bit and grey scale images. They contain hidden information in a same way to watermarks on actual paper. They are referred to as digital watermarks because of this. 
Masking an images changes the image. To ensure that these changes are not detected, they are made in small proportions. The masking and filtering technique usually start with an analysis of the image. Then the significant areas are found, where the hidden message will be more integrated to cover the image and at the end the data is embedded in that particular area. This technique embeds the secret information in more significant areas rather than hiding it only in the noisy section.
Our academic experts are ready and waiting to assist with any writing project you may have. From simple essay plans, through to full dissertations, you can guarantee we have a service perfectly matched to your needs.View our services
This method extends the data of an image by masking the secret data over the original data as opposed to hiding the information inside of data. This causes some arguments among experts that it is definitely a type of Information Hiding,but may be not technically Steganography. In this paper it is considered it to be a steganography technique. Masking and Filtering techniques are that they are immune to image manipulation. While masking do change the visible property of the image, it can be implemented in a way that the human eye will not notice the anomality.
Comparing to LSB, masking is a more robust technique as they are more immune to image manipulation.These masked images passes cropping, compression and also some amount of image processing. Masking embed the information in significant areas therefore the hidden message is more integral to the cover image than it being hidden in the noise level. Thus it is better than LSB for example lossy JPEG images. The poor quality of recover image is the main drawback.
This section covers some well known and important steganalysis techniques that have been developed and helps in understanding the process better. The chisquare attack is a very simple and famous method to test the robustness of the security system against the attacks. Chi-square attack is effective to LSB steganography for spatial images as well as JPEG images.A chi square test [9,14] is used to detect hidden messages by making use of the pairwise dependency concept. It is dependent on probability analysis of stego image after information is put inside it using LSB method then it is compared with the probability analysis of the original image to see the difference between them. If the difference happens to be near zero then it means no information inside the image otherwise it means the image held the information. It is quite relaible for sequentially embedded messages. In further future works it was improved to detect randomly scattered secret messages.
Fridrich et al. developed Raw Quick Pair (RQP) for 24 bit colour images . He later improved this for both grey scale and colour images. Fridrich et al.  putforth the RS algorithm for steganalysis to detect LSB embedding in 8 bit grey scale images and 24bit color images. In this technique the total number of image pixels are divided into four pixels (2×2 blocks) disjoint groups. Discriminating function f (.) is applied to these groups to capture the smoothness of the groups.
f (G)=f(x1,x2,……,xn) = : x1,x2,……,xn are pixels of group G.
Also, three invertible function operations are used on pixel value x.
F1(x): 0 ↔ 1, 2 ↔ 3… 254 ↔ 255
F-1(x): −1 ↔ 0, 1 ↔2… 255 ↔ 256
F0(x): Identity function
Operation F1(x) and F-1(x) are applied on the group of pixel based upon Mask M which is an n-tuple that has values -1,0,1. For eg, if the values of the four pixels of a group G are 13, 23, 46, 64 and M = (1, 0, 1, 0), then FM (G) = (F1 (13), F0 (23), F1 (46), F0 (64)). The defined function, mask and operation areapplied to group of pixels and distributed them into following groups:
Regular groups: G ∈ R ⇔ f (F (G)) > f (G)
Singular groups: G ∈ S ⇔ f (F (G)) < f (G)
Unusable groups: G ∈ U ⇔ f (F (G)) = f (G)
This method is based upon analyzing how the number of singular and regular groups changes with the increase in the message length embedded in the LSB plane. The bigger the message size, the lower the difference between R-M and S-M and the bigger the difference between and RM and SM. This behavioral variation is used in detection of hidden message from the stegoimage.
Westfeld et al.  proposed a technique based upon the statistical analysis of Pairs of Values (PoVs) using chi-square attack that is exchanged during the message embedding. The idea of statistical analysis is to compare the expected frequency theoretically (which is the arithmetic mean of two frequencies in a PoV since the original cover medium is not available) distribution in steganograms with some sample distribution of stego medium. To get the statistics for the difference between distributions of
PoV’s used the below mentioned equation:
The probability of embedding is decided by calculating the p-value defined below:
This p-value is calculated for any sample from the values examined which usually starts from the beginning of the image and gets amplified for each measurement.
This method gives reliable results when the message placement is known to us (for example, sequential).But randomly scattered messages can be reliably detected with this method only when the the number of pixels in the image is comparable with message length.
Dumitrescu  inspired by Fridrich’s work invented LSB based steganalysis technique.It works on the basis of a finite state machine. The operations are based on the behaviour of trace multi sets. Later this method was improved by analyzing the joint and marginal probabilistic distributions of image. Another variation to it was by using least square estimation which led to improved estimation accuracy.  Another improvement was made on this method by deriving detection and estimation equations from higher order statistics. This technique was effective and robust on both colour and grey scale images. .
Zhang and Ping developed another method using statistical histograms to identify weak correlations between LSB and other planes and thus distinguishing clean images from stego images. Exact length of the embedded message can also be detected by this technique. . A method called Pair Analysis Technique was Proposed by Fridrich et al. for palette images. It was basically for 8 bit GIF images. Chen and Chang devised a method that works well on LSB embedded images. It is based on bit plane randomness test and the image is anlayzed by Hilbert scan. These are some of the steganalysis techniques that are most referenced.
From the information collected from forum discussions, the three most used steganographic packages in olden times are StegoDoS, White Noise Storm and S-Tools for Windows to conceal one text message and one image file using two different cover files. A brief description of the software packages follows:
• StegoDoS: It makes use of LSBs to hide messages. It is the least successful out of the three tools. It also appends a end of file character to the end of the message. The message retrieved from the altered image is most likely to contain garbage at the end even with the EOF character.
• White Noise Storm: It is a very effective steganography application for DOS. It includes an encryption routine to randomize the bits inside an image. The software uses the LSB approach and applies this method to IBM Paintbrush (PCX) files. The software extracts the LSBs from the cover image and they are stored in a file. Then the message is encrypted and applied to these bits to create a new set of LSBs. To create the new stego-image, the modified bits are then injected to the cover image.
• S-Tools for Windows: S-Tools can apply the LSB method to both images and audio files. It is the most versatile steganography tool out of the three due to this property. Version 3 also includes programs that process GIF and BMP images and audio WAV files. S-Tools can also be used to hide information in the “unused” areas on floppy diskettes. Where as version 4 incorporates image and sound file processing into a single program. In addition to supporting 24-bit images, S-Tools also includes encryption routines with varied options. S-Tools undoubtedly provided the best results for steganography among the three tools.
Some other tools are :
•Jsteg – It embeds information in JPEG image format. It hides secret image in the GIF by replacing the LSBs of quantized run lengthy coded DCT coefficients with secret message bits during the time of JPEG compression. The message embedding using Jsteg introduces ststistical signature that can be used for attack against JSteg
•JP Hide and Seek It embeds a secret message in colour JPEG images. It embeds bits by replacing LSBs of the quantized DCTcoefficients during JPEG compression. Like previous case statistical signature is an issue here.
In this paper various image steganography techniques and steganalysis methods have been discussed. When compared with chi square method[9,14] ,RS is more relaible.( RS is unable to detect messages that are less than 0.005 bpp). When raw loseless compressed images are the analyzed format of image steganalysis technique put forth by Zhang and Ping is more effective than RS. In the case of palette and BMP images, Pair Analysis Technique is superior than RS. Where as for grayscale images RS works better.
In todays day and age, ‘Hacking’ is a term often used. Hacking is an unauthorized access of data which is collected at the time of data transmission without the permission of the owner. In case of steganography this issue is often taken as Steganalysis. Steganalysis is a process by which a steganalyzer cracks the cover object to get the data that is hidden. Therefore whatever is the technique that is developed in future, degree of security of that is is high and needs to be kept in mind. It is possible that Dual Steganography or Steganography along with Cryptography may be the future solution for this above mentioned issues.
 H.-T. Wu and J.-W. Huang, “Secure JPEG steganography by LSB+ matching and multi-band embedding,”,18th IEEE International Conference on Image Processing 2011
 C. K. Chan and L. M. Chang, “Hiding Data in Images by Simple LSB Substitution,” Pattern Recognition, Vol.37, pp.469-474, 2004.
 H. J. Zhang and H. J. Tang, “A Novel Image Steganography Algorithm Against Statistical Analysis,” Proceedings of Sixth International Conference on Machine Learning and Cybernetics, pp.3884-3888, 2007.
 Robert Krenn, Steganography and steganalysis, Internet Publication, March 2004. Available at: http://www.krenn.nl/univ/cry/steg/article.pdf (Last accessed on 5/5/’19 at 3.42pm)
 T. Pevny and J. Fridrich, “Determining the stego algorithm for JPEG images,” Proc. Inst. Elect. Eng., Inf. Security, vol. 153, no. 3, pp. 77–164, 2006.
 A. D. Ker, “Steganalysis of LSB matching in grayscale images”, IEEE Trans. Signal Process, vol. 12, no. 6, pp. 441-444, Jun. 2005
 J. Fridrich, M. Goljan and R. Du, “Reliable Detection of LSB Steganography in Color and Grayscale Images”. Proc. Of ACM Workshop on Multimedia and Security, Ottawa, Oct. 5, 2001, pp. 27-30
 A. Westfeld, A. Pfitzmann, “Attacks on Steganographic Systems”, in Andreas Pfitzmann (Ed.): Information Hiding, LNCS 1768, pp. 61-76, Springer-Verlag, 1999.
 Somnath Maiti, Manas Ranjan Nayak,Subir Kumar Sarkar, “Modified Least Significant Bit(LSB) Matching Technique for Robust Information Hiding”, JETIR (ISSN-2349-5162), September 2017, Volume 4, Issue 09
 Bender, W., D. Gruhl, N. Morimoto, & A. Lu, 1996. Techniques for data hiding. IBM Systems Journal, 35(3.4): 313-336. doi: 10.1147/sj.353.0313
 P. Vignesh Pejathaya, Imran Khan, Vighnesh Shetty, Sayeesh” Review on High Security Digital Data Steganography for Secret Communication”, International Journal of Engineering Research & Technology (IJERT), Vol. 8 Issue 02, February-2019.
 Handbook of Research on Secure Multimedia Distribution Textbook by Shiguo Lian 2009 1st edition
 Fridrich J,GoljanM(2002) Practical steganalysis of digital images-state of the art, in: Proc. SPIE PhotonicsWest, Electronic Imaging, Security and Watermarking of Multimedia Contents, San Jose, CA, vol. 4675, 1–13
 Westfeld A (2002) Detecting low embedding rates. in: Lecture Notes in Computer Science, vol. 2578,Springer-Verlag, Berlin, 324–339
 Fridrich J, Du R, Meng L (2000) Steganalysis of LSB encoding in color images. in: Proc. IEEE Int. Conf.on Multimedia and Expo, New York, July 31–August 2, 2000
 Dumitrescu S, Wu X, Wang Z (2003) Detection of LSB steganography via sample pair analysis. IEEE
Trans. Signal Process, 1995–2007
 Lu P, Luo X, Tang Q, Shen L (2004) An improved sample pairs method for detection of LSB embedding. in: Proc. 6th Int. Workshop Inf. Hiding, Toronto, ON, Canada, May 2004
 Dumitrescu S, Wu X (2005) A new framework of LSB steganalysis of digital media. IEEE Trans Signal
 Zhi L, Fen SA, Xian YY (2003) A LSB steganography detection algorithm. in: Proc. IEEE Int. Symp. On Personal, Indoor and Mobile Radio Communication
 Fridrich J, Goljan M, Soukal D (2003) Higher-order statistical steganalysis of palette images. in: Proc. SPIE,Electronic Imaging, Security, Steganography, and Watermarking of Multimedia Contents V, Santa Clara, CA, vol. 5020, 178–190
 Saman Shojae Chaeikar1 & Mazdak Zamani1 &
Azizah Bt Abdul Manaf1 & Akram M. Zeki, “PSW statistical LSB image steganalysis “, Springer January 2017
Cite This Work
To export a reference to this article please select a referencing stye below: