Why Image Encryption for Better Steganography

RESULTS AND DISCUSSION

A perfect encryption algorithmic rule should stand for it in opposition to Cryptanalysis assails. Contrasting the conventional encryption methods, the requisites of digital encryption sound dissimilar. It does need cryptographic together with perceptual refuge. In this script, algorithm is run in MATLAB 7.1 with gray Lena as plain images which are of dimension 256x256 and their shuffled, encrypted versions and their corresponding histograms are shown in Fig. 3a-f, respectively. Tentative results for this study are depicted in Table 1.

In this script, algorithm is run in MATLAB 7.1 with gray Baboon as plain images which are of dimension 256x256 and their shuffled, encrypted versions and their corresponding histograms are shown in Fig. 4a-f, respectively.

In this script, algorithm is run in MATLAB 7.1 with gray Mahatma gandhi as plain images which are of dimension 256x256 and their shuffled, encrypted versions and their corresponding histograms are shown in Fig. 5a-f, respectively.

Table 1:	Experimental results of image metrics obtained for image encryption

Fig. 3(a-f):	(a) Lena, (b) Secret image shuffled image, (c) Encrypted image, Histogram of (d) Secret image, (e) Shuffled image and (f) Encrypted image

In this script, algorithm is run in MATLAB 7.1 with gray Kovil as plain images which are of dimension 256x256 and their shuffled, encrypted versions and their corresponding histograms are shown in Fig. 6a-f, respectively.

Stochasticity tests: Ergodic properties top the priorities of a consummate encryption model. Original and enciphered images are shown in figures which look absolutely absurd. Random literally mentions the processes bring forth identically detached and free-lance samples. The process is random if pragmatic value is determined through its position in sequence else through subsequent observations. Tentative outcomes suggest that this output produced in this paper is absolutely random giving no hint about the clandestine information.

Fig. 4(a-f):	Baboon (a) Secret image, (b) Shuffled image, (c) Encrypted image Histogram of (d) Secret image (e) Shuffled image (f) Encrypted image

Histogram tests: Histograms for pixels in all the four images are given. It is noticed that plain image histograms demonstrate a pattern whereas pixels distribution in encrypted version is flat. It signifies the fact that frequency investigation is unable to crack this algorithm. Also, in order to not letting the intruders to know about the covert information, it is mandatory that the histograms of original and end result remain statistically identical. Thus, in this paper encrypted histograms are appreciably unlike from their original ones. Plain histograms have huge cum spiky peaks trailed by jagged declines but the resultants comprise of homogenous variations.

Correlation of pixels: As the final outputs give the impression of being unsystematic, it is okay if third party expects zero correlation. As a matter of fact, pixel correlation is a key measure of performance in any image encryption algorithm. Here presented the values of all horizontal, vertical and diagonal correlation among pixels in both plain and encrypted images.

Fig. 5(a-f):	Mahatma Gandhi, (a) secret image (b) shuffled image (c) encrypted image Histogram of (d) Secret image (e) Shuffled image and (f) Encrypted image

While for the former the correlation is nearly equal to 1 and almost 0 is the case for the latter. Thus plain images exhibit high correlation and encrypted images depict negligible correlation between adjacent pixel pairs in this study. The correlation coefficient is expressed by:

where, σ = i and j’s standard deviations, respectively cov(i, j) = Covariance of i and j.
Covariance is given as:

Fig. 6(a-f):	Kovil: (a) Secret image (b) Shuffled image (c) Encrypted image Histogram of (d) Secret image (e) shuffled image (f) Encrypted image

Here, μ_i and μ_j are i and j’s mean, respectively.

Differential attack: This index is to quantify the security of the algorithm. Here by altering a single pixel one can observe the subsequent resultant image. If there is any such change, then this run is said to be bungling and not viable. The two constraints involved in studying persuasion of a solitary pixel change are NPCR and UACI. Their expressions are given below:

This study is differential attack tested by exposing all the images which are of varying size. NPCR of approximately 99.6% and UACR of roughly 33.6% are witnessed in this paper which is fairly upright. Thus one cannot trace a small similarity between the original and resultant images. As a result, differential attack is not viable on this algorithm.

Image metrics: Making it ideal from the rest, this study is vindicated by BER, MSSIM and PSNR. These images have very less PSNR which signifies the fact that there is no relation between the original and the resultants. Moreover, the third party cannot predict that there is some confidential information contained in them. In simple, PSNR predicts the resemblance between two images which in this case is anticipated.

Mssim generally lies between -1 and 1; the nearer the value to 1 higher is the similarity between the images. The proposed model exhibits MSSIM of almost 0 that makes the paper take pride in security grounds. This means only trifling blunder is encountered by this algorithm which is for sure allowable and acceptable.

Where:

μ_O is the estimate of the mean intensity of the cover (N = 255), σ₀ is the standard deviation

Here σ_OS is correlation coefficient.

BER is also a performance criterion which tells about the error witnessed in the procedure. This proposal gives it approximately as 0.5 for all the images. It confirms there are almost 50% error rate.

CONCLUSION

This study introduces an inimitable image encryption model which takes Cipher Block Chaining Mode as the fundamental element and constructs further plots to achieve a secret sharing scheme. The algorithm takes advantage of dissemination and mystification via spoofing the secret image and with the help of keys and scrambling, the relation between the secret images and encrypted images is brought to zilch. The paper withstands some notable attacks namely differential, correlation, histograms to prove its creation for which simulation upshots are illustrated. It maximizes the efficiency by significantly improving the randomness among pixels and security as well. Experimental results confirm the practicality of this proposal. These features boost to put forward the paper as a pertinent and promising encryption algorithm. Since it is an image encryption routine, it does not bestow data expanding and thus is very much sensible to parameters having a good prognosis in information security.