Page 179 - 2024-Vol20-Issue2

175 | Muttashar & Fyath

TABLE IV. TABLE V.
PSNR VALUE BEFORE AND AFTER APPLIED DNCNN COMPARISON OF ENTROPY WITH RELATED WORK FOR
WITH DIFFERENT STANDARD DEVIATION s OF GAUSSIAN
DOUBLE AND MULTIPLE IMAGES.
NOISE.
Ref. Encryption No. of Chaotic Entropy
Scheme Image System
Name Before DnCNN Gray- 5D 7.9991
of double
s = 0.5 s = 5 s = 10 s = 15 s = 20 [15] Hybrid 2D-Logisic R-7.7358
Images Color- Map G-7.6303
Lena 27.01 17.28 15.17 13.91 13.04 [5] Optical single B- 7.5681
Baboon 23.80 15.06 12.97 11.83 11.12 Hyperchaotic R-7.9974
Barbara 26.97 17.37 15.25 13.98 13.17 [28] Digital Color- G-7.9971
Airplane 23.95 14.57 12.25 11.22 10.55 single Memristive B-7.9973
Sailboat 27.05 17.24 14.93 13.63 12.74 [2] optical chaos R-7.9900
Peppers 23.67 14.54 12.36 11.27 10.54 Color- G-7.9969
After DnCNN This Hybrid single 9D B-7.9972
Lena 32.56 23.72 21.34 19.62 18.48 Work R-7.9991
Baboon 21.86 19.51 17.88 16.79 16.02 Color- G-7.9987
Barbara 27.91 23.25 21.32 19.90 18.90 multiple B-7.9991
Airplane 28.13 21.80 17.96 16.17 15.01
Sailboat 27.65 22.82 20.76 19.15 17.87 VII. CONCLUSIONS
Peppers 29.24 20.64 17.16 16.11 15.19
A 9D chaotic-based hybrid digital / optical encryption scheme
denoising by DnCNN for noisy images with different values has been proposed for double-color images. Each of the three
of standard deviation s . MATLAB R2021b is used to train chaotic sequences has been used to control the encryption of
the denoising model. Fig. 26 shows examples of original one of the RGB channels without affecting the other channels.
peppers images, noisy, and denoised image for different noise One chaotic sequence controls the channel’s fusion, XOR,
values. The parameter values of the training options are as and scrambling-based digital encryption part. The other two
follows. chaotic sequences have been used to construct two chaotic
phase masks implemented in the optical FT-based encryption
i. The learning rate is 110-4. part. A deep learning technique has been used to reduce the
effect of Gaussian noise embedded in the received encrypted
ii. The training network sets initial weights. images. The simulation tests reveal that using a 9D chaotic
system to control the operation of the HDOE scheme, with
iii. The number of epochs is 10 for the trained model. each three sequences, are responsible for one of the RGB
channels, increases the security level and robustness of the en-
VI. COMPARATION WITH RELATED WORK cryption. The scheme offers 7.9989 entropy for the encrypted
color images and an infinite PSNR for the decrypted images.
This section gives a brief comparison of encryption measures The encryption algorithm successfully resists different attacks,
obtained by using the proposed encryption scheme with that such as noise and cropping attacks. Further, The designed
reported in related work. The three measures used for com- DnCNN operates efficiently with the proposed encryption
parison are entropy, encryption key sensitivity, and key space. scheme yielding performance enhancement of the decrypted
For entropy comparison, the results of this work are compared images against Gaussian noise. The next step in this work is
with that of Refs. [2, 5, 15, 28].The comparison results are to modify the scheme to deal with more than two color images
listed in Table V along with brief description parameters of by adopting compressive sensing techniques.
the used encryption scheme. Investigation of the results of
this table reveals that the proposed scheme in this work offers CONFLICT OF INTEREST
an entropy that is nearest to the ideal case (entropy = 8) com-
pared with other references. The authors have no conflict of relevant interest to this article.
The proposed scheme offers 10-17 key sensitivity and 2903
key space. These results indicate that the proposed scheme has
higher key sensitivity and a much higher key space dimension.