Some features of a steganography method:
Embedding capacity - An embedding capacity too small would not be useful to fit a secret message. Images and PDFs are the most popular cover objects for their large embedding capacity.
Imperceptibility - This measurement is required because these methods produce changes in pixel values, image histograms, or generate noise in stego images and it is necessary to keep distortions minimal.
Undetectability - If a steganalysis algorithm detects the inspected media as the stego one, even without knowing any extra information about the secret message, the steganographic approach fails. This is in contrast to encryption, where a successful attack means that the attacker gains access to the decrypted content or partially recovers the encryption key.
Some metrics used to compare methods:
Image histograms - Histogram shift can occur due to changes in pixel values due to the embedding of messages. We can use image histograms to measure the quality of a method[1].
PSNR - Peak signal to noise ration - The ratio between the maximum possible power of a signal and the power of corrupting noise that affects the fidelity of its representation.
SSIM - Structural similarity index - Luminance, contrast, and structure replace the summation method used as the basis for calculating PSNR.
Grayscale vs Color images - Steganographic methods on grayscale images and color images have different effects, in a research conducted [1] the same comparison is done on the grayscale baboon image, the result is PSNR and SSIM agree that the LSB method is better, while the color baboon image produces different values.
Spatial vs transform domain steganography - The transform domain is superior in terms of resistance to manipulation so it is more suitable for use in image watermarking because it can maintain the data embedded in the image container from manipulation damage [1][2][3][4]. Steganography is more focused on imperceptibility, payload capacity, and security.
LSB is a method that is very simple and superior in terms of imperceptibility and payload, this is because by default this method only makes a very small change of 1-bit in each pixel of the image container[5][6][7].
In some research [8][9] LSB method was developed to increase the security and payload capacity. PSNR measurement results for the method proposed in the two studies are indeed quite good, namely above 50 dB, but from the histogram presented it appears that there is a histogram sample that changes visibly. This makes this method less good and can cause suspicion of the attacker
LSB is focused on making the embedding imperceptible rather than undetectable, but for a steganographic method to "fail" it is only required that the embedding is detected.
When altering LSB bits to fit a message, the bits are changed on an average 1/2 of the time. When the LSB of a value is changed, the odd value always decreases and the even value always increases.
This alters the image histogram making the steganography detectable but SPA (Sample pair analysis) is a reliable and better[10] attack that detects LSB.
The embedding operation of flipping the LSB creates many problems due to its asymmetry (even values are never decreased and odd values never increased during embedding).
When an LSB needs to be changed, instead of flipping the LSB, the value is randomly increased or decreased, with the exception that the values 0 and 255 are only increased or decreased.
The extraction algorithm of ±1 embedding is the same as for LSB embedding – the message is read by extracting the LSBs of cover elements.
Sample pairs analysis fails on ±1 embedding.
Detecting LSBM: ±1 embedding in the spatial domain can be detected using the center of gravity of the histogram characteristic function (absolute value of the Fourier transform of the histogram) as the feature. This is because adding an independent stego noise to the cover image smooths the histogram. The accuracy of this method can be improved by considering the adjacency histogram and by applying calibration (resampling the stego image). [11]
| [1] | Aini DN, Setiadi DRIM, Putro SN, et al (2019) Survey of methods in the spatial domain image steganography based imperceptibility and payload capacity |
| [2] | Cheddad A, Condell J, Curran K, Mc Kevitt P (2010) Digital image steganography: survey and analysis of current methods. |
| [3] | Hussain M, Wahab AWA, Bin IYI et al (2018) Image steganography in spatial domain: a survey. Signal Process Image Commun 65:46–66. https://doi.org/10.1016/j.image.2018.03.012 |
| [4] | Setiadi DRIM (2019) Improved payload capacity in LSB image steganography uses dilated hybrid edge detection. J King Saud Univ - Comput InfSci https://doi.org/10.1016/j.jksuci.2019.12.007 |
| [5] | Setiadi DRIM (2019) Payload enhancement on least significant bit image steganography using edge area dilation.https://doi.org/10.24425/ijet.2019.126313 |
| [6] | Chakraborty S, Jalal AS, Bhatnagar C (2017) LSB based non blind predictive edge adaptive image steganography. Multimed Tools Appl 76:7973–7987. https://doi.org/10.1007/s11042-016-3449-4 |
| [7] | Muhammad K, Ahmad J, Rehman NU, Jan Z, Sajjad M (2017) CISSKA-LSB: color image steganography using stego key-directed adaptive LSB substitution method. Multimed Tools Appl 76:8597–8626.https://doi.org/10.1007/s11042-016-3383-5 |
| [8] | Astuti YP, Setiadi DRIM, Rachmawanto EH, Sari CA (2018) Simple and secure image steganography using LSB and triple XOR operation on MSB. In: 2018 International conference on information and communications technology, ICOIACT 2018. Yogyakarta |
| [9] | Setiadi DRIM, Santoso HA, Rachmawanto EH, Sari CA (2018) An improved message capacity and security using divide and modulus function in spatial domain steganography. In: 2018 international conference on information and communications technology (ICOIACT). IEEE, Yogyakarta, pp 186–190 |
| [10] | S. Dumitrescu, X. Wu, and N. D. Memon. On steganalysis of random LSB embedding in continuous-tone images. In Proceedings IEEE, International Conference on Image Processing, ICIP 2002, pages 324–339, Rochester, NY, September 22–25, 2002 |
| [11] | Harmsen and W. A. Pearlman. Steganalysis of additive noise modelable information hiding. In E. J. Delp and P. W. Wong, editors, Proceedings SPIE, Electronic Imaging, Security and Watermarking of Multimedia Contents V, volume 5020, pages 131–142, Santa Clara, CA, January 21–24, 2003. |